The PEGS DREAM Challenge: A Crowdsourcing Approach to Understanding Hypercholesterolemia with Multi- dimensional Genomic and Environmental Data

The PEGS DREAM Challenge: A Crowdsourcing Approach to Understanding Hypercholesterolemia with Multi- dimensional Genomic and Environmental Data | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article The PEGS DREAM Challenge: A Crowdsourcing Approach to Understanding Hypercholesterolemia with Multi- dimensional Genomic and Environmental Data Farida S. Akhtari, Johannes Falk, Jyoti Jyoti, Venetia Voutsa, and 15 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9254914/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 7 You are reading this latest preprint version Abstract Crowdsourced challenges are powerful catalysts for advancing biomedical research and fostering community-driven innovation. The DREAM Challenges initiative, in collaboration with the Personalized Environment and Genes Study (PEGS), held a competition to spur the development of predictive models for hypercholesterolemia risk. Participants were tasked with integrating diverse data, including environmental exposures, whole-genome sequencing, and geospatial information, from a large and diverse cohort to classify hypercholesterolemia and generate novel insights. The top-performing models, which primarily leveraged gradient boosting and random forest classifiers, demonstrated strong predictive performance, outperforming traditional polygenic scores (PGS). In Challenge Task 1, models from the two top-performing teams substantially outperformed the Challenge benchmark dataset using only PGS (AUROC = 0.7358), with AUROCs of 0.7933 and 0.7919, respectively. Beyond prediction, these models highlight the significant value of self-reported health and environmental exposure data, revealing them as informative predictors of hypercholesterolemia risk that complement established clinical and genetic factors. This paper details the design of the challenge, presents the top-performing models, and highlights the potential of integrative, multimodal approaches for understanding and predicting complex human diseases. Biological sciences/Computational biology and bioinformatics Biological sciences/Genetics DREAM Challenge Personalized Environment and Genes Study gene-environment interactions hypercholesterolemia high cholesterol crowdsourced data analysis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Recognizing the profound influence of gene-environment (GxE) interactions on human health and disease, researchers are increasingly integrating genetic and environmental data in studies of etiology. Hypercholesterolemia, commonly called high cholesterol, is a highly prevalent condition and carries a significant risk for cardiovascular disease 1 . This condition exemplifies the complexity of GxE interactions and their impacts on disease 2 – 4 . While genetic predisposition contributes to individual susceptibility to high cholesterol, lifestyle choices, dietary habits, and environmental exposures also influence hypercholesterolemia status. Traditional approaches often analyze individual data types in isolation, limiting their ability to capture the full spectrum of disease-driving factors. The recent focus on high-dimensional datasets containing multi-omics data, or data from multiple biological systems, is enabling researchers to elucidate the many factors influencing complex conditions such as hypercholesterolemia. For example, incorporating information on environmental factors with genomic data can reveal how risk alleles interact with specific environmental exposures to modulate disease risk. Integrating diverse data types supports a holistic understanding of disease mechanisms, paving the way for more accurate risk prediction, targeted interventions, and personalized medicine. Dialogue for Reverse Engineering Assessment and Methods (DREAM) Challenges ( https://dreamchallenges.org/ ), in partnership with Sage Bionetworks, are aimed at translating the “wisdom of the crowd” into practice with potentially significant impacts on science and human health. The Challenges engage the scientific community in collaborative problem-solving to address fundamental biomedical issues such as disease classification and prediction using innovative computational models. These global Challenges bring together researchers and data scientists from multiple disciplines to develop and benchmark informatic algorithms. In the DREAM Challenge framework, teams address a specific research question by preparing relevant data and developing models, which are then evaluated and shared. DREAM Challenges have addressed a range of diseases, and results have been published in numerous academic journals 5 . A DREAM Challenge using data from the Personalized Environment and Genes Study (PEGS), a diverse North Carolina-based cohort of nearly 20,000 individuals, was launched to promote innovative statistical and data science approaches integrating multi-dimensional environmental, genomic, and geospatial data to dissect the etiology of hypercholesterolemia (Fig. 1 ). The PEGS cohort is described in detail on the PEGS website ( https://www.niehs.nih.gov/research/atniehs/labs/crb/studies/pegs ). Exposomic data gathered through comprehensive surveys capture PEGS participants’ endogenous and exogenous exposures throughout life, including chemical and environmental exposures encountered at work and home, medications, dietary habits, and lifestyle choices. This comprehensive environmental data enables researchers to consider exposures’ cumulative effects and interactions with genomic factors. The whole-genome sequencing (WGS) data available for the PEGS cohort comprises single nucleotide variants (SNVs), structural variants, human leukocyte antigen (HLA) genotypes, telomeric content, ancestry estimations, and genome-wide methylation profiles (Fig. 2 ), offering insights into both common and rare genetic variations and their potential functional consequences. Geospatial data were created by linking PEGS participants’ addresses to databases that include proximity to contaminant sources, area-level air pollutant concentrations, and social determinants of health indices such as the CDC/ATSDR Social Vulnerability Index (SVI) ( https://www.atsdr.cdc.gov/placeandhealth/svi/index.html ) and the Environmental Justice Index (EJI) ( https://www.atsdr.cdc.gov/placeandhealth/eji/index.html ) (Fig. 3 ). These indices provide valuable insights into the spatial distribution of environmental risks and their impacts on health disparities. The PEGS DREAM Challenge ( https://www.synapse.org/PEGS ) asked participants to develop novel predictive models for hypercholesterolemia by harnessing the power of this multi-omics dataset, with the aim of more accurate disease classification and risk prediction and improved ability to develop targeted interventions and inform personalized medicine. The Challenge featured distinct classification and ideation tasks. In the classification task (Task 1), teams developed models in the multi-omics PEGS data to classify hypercholesterolemia. To assess the potential value of incorporating exposomic and geospatial data for improving classification accuracy, a polygenic score (PGS) derived solely from WGS data was used as a benchmark. For the ideation task (Task 2), teams generated data-driven hypotheses and/or models on the relationships between the multidimensional PEGS data and high cholesterol risk. The ideation task incentivized outside-the-box thinking and promoted collaboration, with the overall aim of unraveling how GxE interactions affect human health. Results PEGS DREAM Challenge Data The PEGS DREAM Challenge (https://www.synapse.org/PEGS) asked participants to leverage the multi-dimensional PEGS dataset to identify factors affecting hypercholesterolemia and thus improve understanding of its etiology. The challenge was launched on May 1, 2024, with a Leaderboard round from May 10 to August 6, 2024, and a Final round from August 6 to August 19, 2024. Ninety participants from five continents—Asia, Africa, Europe, North America, and Australia—completed 345 submissions across all rounds of the Challenge. Winners were announced on August 31, 2024 and presented their findings at the Regulatory & Systems Genomics with DREAM Challenges Conference (RSGDREAM2024) in Madison, Wisconsin, USA (https://www.iscb.org/rsgdream2024/home). Participants took part in one or both tasks: Task 1: Disease Classification: Classify individual hypercholesterolemia disease status in a held-out test subset containing health, exposomic, geospatial, and genomic data. Task 2: Ideation Challenge: Develop novel hypotheses and models using the multi-dimensional PEGS data to improve understanding of hypercholesterolemia etiology beyond conventional clinical and genetic risk factors. Participants were provided questionnaire-based health and exposure, geospatial, and genomic data from PEGS Data Freeze 3.1. Table 4 provides details. Questionnaire-based data: Demographic, health, environmental exposure, socioeconomic status, and lifestyle data collected from three surveys administered to PEGS participants: the Health & Exposure Survey (N = 9,449), the Internal Exposome Survey (N = 3,071), and the External Exposome Survey (N = 3,618). Genomic data: WGS data (N=4,737), including SNVs and indel genotypes, structural variant calls, HLA genotypes, aggregate telomeric content, ancestry estimations, and genome-wide methylation profiling data (N=4,724). Methylation profiling was done using the Infinium MethylationEPIC v1.0 BeadChip Kit. Geospatial data: Exposure estimates and proximity to hazards calculated using geospatial linkages (e.g., proximity to contaminant sources, air pollutant concentrations), Modern Era Retrospective analysis for Research and Applications (MERRA-2) climate/environmental estimates, and SVI and EJI data. The Challenge data (N = 9,449) were filtered to remove outliers, related individuals, and race mismatches, resulting in a dataset comprising 9,184 individuals. The filtered data were split into individuals with and without WGS data. Each subset was then randomly split into thirds to create training (n = 3,062), validation (n = 3,062), and test (n = 3,060) subsets. Each subset comprised approximately 50% individuals with sequencing data and 50% without sequencing data. The test subset was not shared with Challenge participants and was held out to score Final round submissions. Figure 4 shows the workflow for Task 1. To protect the privacy of PEGS participants, the original PEGS data were not provided to Challenge participants. Synthetic data, created using the synthpop R library, was provided for model construction and code development. The synthetic data mirrored the structure and sample sizes of the original dataset but did not preserve inter-table correlations. Submitted models were evaluated with the original PEGS data. Privacy measures included de-identification and anonymization with generated IDs in addition to limiting access to original data. Challenge Task 1: Disease Classification In Task 1, participants developed models in the training and validation datasets to classify hypercholesterolemia status in the held-out test dataset. Participants were encouraged to surpass a benchmark area under the receiver operating characteristic curve (AUROC) of 0.7358, obtained from a hypercholesterolemia PGS. This benchmark PGS was computed for PEGS participants using weights and metadata from the hypercholesterolemia PGS by Weissbrod et al. 6 in the Polygenic Score Catalog 7,8 . This PGS was selected due to its inclusion of multiple ancestries and a large number of genomic variants. Model accuracy was assessed using AUROC, with area under the precision-recall curve (AUPRC) as a tiebreaking metric. Teams Spider Bobs (AUROC = 0.7933) and Nonsense-Mediated Decay (AUROC = 0.7919) submitted the top-performing submissions for Task 1. Team Spider Bobs (Task 1) Team Spider Bobs integrated data from the Health & Exposure Survey and Internal Exposome Survey using random forest and gradient boosting classifiers from Python’s scikit-learn library 9 . For data processing, some variables were manually preprocessed. For ordinal variables, coded responses were reordered to reflect ordinality (e.g., Health & Exposure Survey - current physical health rating compared to five years ago as better, worse, or about the same). For continuous (height, weight, age, BMI) and some ordinal (Health & Exposure Survey questions in the fatigue section) variables, missing responses were imputed with the mean value. For skipped questions (e.g., smoking details for non-smokers), missing responses were replaced with appropriate values (e.g., 0). Gradient boosting had better performance on the team’s cleaned data and was used for the final model. The team optimized hyperparameters using cross-validation. Reducing the number of features and sub-samples significantly improved predictions. Team Spider Bobs used the following workflow (Figure 5): Benchmark Model (Figure 5a): An initial model using key features from the Health & Exposure survey (weight, height, sex, age) and trained with a multi-layer perceptron (MLP) had an AUROC of 0.7356 in the validation data. Refinement (Figure 5b): The model was refined by leveraging PEGS exposome-wide association studies (ExWAS) results and correlation globes (https://pegsexplorer.niehs.nih.gov) to identify relevant features based on P -values and correlations 10,11 . Features were selected based on ExWAS P -values and odds ratios, and correlations were checked using correlation globes. For each survey data component outlined in Table 1 (except demographic and administrative data), features were connected if the correlation value was > 0.6, and a representative feature from each component was selected (Figure 5). These were further reduced to three main features plus benchmark features to mitigate overfitting. Data were partitioned based on the availability of Internal Exposome Survey data, and two MLP models were trained and applied. This refined approach improved the AUROC to 0.7412. Final Model (Figure 5c): For feature selection to identify the most significant features for the final model, a literature review, data processing, and hyperparameter optimization were utilized. Data were split again, and two gradient boosting classifiers were trained. Team Nonsense-Mediated Decay (Task 1) Team Nonsense-Mediated Decay integrated Health & Exposure Survey and genetic data to enhance classification accuracy. Decision trees and ensemble methods such as random forest and XGBoost were used to capture non-linear relationships between mixed data types (e.g., numerical and categorical variables) and perform feature selection. For genomic data, PGS were utilized to complement the random forest model and capture the combinations of variants influencing disease risk. Data were processed to limit analysis to numeric and categorical variables from the Health & Exposure Survey. Variables with more than 5% missing values were removed; remaining missing data were imputed using the median for numeric data and mode for categorical data. Team Nonsense-Mediated Decay used the following workflow for a random forest classifier with Health & Exposure Survey data combined with PGS: (Figure 6): Benchmark Model: An initial model was trained with a random forest classifier using only Health & Exposure Survey data to estimate baseline probabilities of hypercholesterolemia status, leveraging the model’s inherent feature selection. Refinement: Twelve PGS associated with hypercholesterolemia were calculated for PEGS participants with genetic data using harmonized weight sets from the PGS Catalog 7 . These PGS and the baseline random forest predictions were combined using logistic regression to refine individual disease probabilities. Final Model: The final probability of hypercholesterolemia for each individual k was computed using the formula: where P̂(Y k ) is the predicted probability of hypercholesterolemia for individual k , X k,b is the baseline probability estimated by the random forest model, X k,gi is the i- th PGS for individual k , and β represents the logistic regression coefficients. This two-step integration optimized and regularized the weight of genetic and survey predictors. The hybrid model achieved improved classification performance (AUROC=0.775 on the validation dataset) with respect to its individual components using survey data or genetic data only (with AUROCs of 0.758 and 0.736, respectively), demonstrating the utility of combining multimodal data. By incorporating PGS, the model leveraged insights from previous large-scale studies to interpret genome-wide variation. Challenge Task 2: Ideation Challenge In Task 2, participants leveraged the multi-dimensional PEGS dataset to develop novel models and hypotheses with the aim of improving understanding of hypercholesterolemia etiology beyond conventional genetic and clinical factors. Judges scored the hypotheses and models based on creativity, significance, interpretability, innovation, utility, feasibility, and potential translational impact. As with Task 1, the top-performing submissions for Task 2 were from Teams Spider Bobs and Nonsense-Mediated Decay . Team Spider Bobs (Task 2) For Task 2, Team Spider Bobs conducted follow-up analyses from their Task 1 model to identify novel factors influencing hypercholesterolemia risk. Methylation data were analyzed using the limma R package 12 to detect significant differentially methylated probes (adjusted P -value < 0.05) for individuals with and without hypercholesterolemia. The results revealed 10 probes (seven hyper, three hypo) corresponding to three hypermethylated genes ( ELOVL2 , FHL2 , ZYG11A ) and two hypomethylated genes ( PXN , CCDC102B ). ELOVL2 is involved in fatty acid elongation, potentially influencing lipid metabolism, and FHL2 has been linked with cardiovascular disease. Next, association analyses were performed for the HLA genotype data using PLINK’s 13 logistic regression model with quality control steps that included removing samples/SNPs with excessive missing data, rare variants, and deviations from Hardy-Weinberg equilibrium. There was a significant difference in the distribution of HLA-DPB1 alleles between individuals with and without hypercholesterolemia (χ² = 60.01, p = 0.0072). This association remained significant after multiple testing correction. Marginally significant associations with HLA-H ( P = 0.0534) and HLA-DMA ( P = 0.0320) were also observed. There is a known association between familial hypercholesterolemia and certain HLA alleles, and elevated total cholesterol has been correlated with specific HLA variants. Additional analysis of features included in the gradient boosting model developed for Task 1 revealed that, in addition to well-known risk factors (e.g., BMI, age, smoking, hypertension, alcohol, poor diet, diabetes), factors related to secondary hypercholesterolemia (i.e., high cholesterol triggered by other diseases), such as uterine tumors, also played an important role. The team identified three novel features significantly associated with hypercholesterolemia status in the PEGS data: Vitamin E taken regularly in the past year (from the Internal Exposome Survey) Regular exposure to dyes (from the Health & Exposure Survey) Ever diagnosed with German Measles (from the Internal Exposome Survey) Using the Informatics for Integrating Biology and the Bedside (i2b2) self-service web-based tool for data exploration (see details in the Methods), Team Spider Bobs further examined these associations. For exposure to dyes, the team hypothesized a correlation with a low standard of living, a known factor for hypercholesterolemia (i.e., workers in chemical/textile manufacturing may earn below-average wages). Surprisingly, the correlation between diagnosis with German Measles and hypercholesterolemia status had an extremely small P -value (p ≈ 6.7e-32). An interaction between the German Measles vaccine and the HLA-DPB1 gene has been reported. A small number of individuals fail to build protective antibodies despite vaccination, and differences in immunity to the Rubella virus have been associated with the HLA-DPB1 gene 14,15 . Based on the observed correlations in PEGS data and improved model scores in Task 1 when including ‘Ever diagnosed with German Measles’, the team hypothesized that infection with German Measles is associated with hypercholesterolemia status in the PEGS data. To test the hypothesis accurately, the team suggested using German Measles antibody results for PEGS participants, as self-reported diagnoses are prone to error (around half of infections go undetected) 16 . This antibody data would enable further testing of the association of hypercholesterolemia and HLA-DPB1 . However, the team acknowledges that this putative association could be due to the potential correlation of German Measles and hypercholesterolemia with socioeconomic factors. Team Nonsense-Mediated Decay (Task 2) For Task 2, Team Nonsense-Mediated Decay proposed a GxE interaction model that integrates genomic data on SNVs with Health & Exposure Survey data. These components were selected due to their broad availability in the cohort and their observed utility in improving Task 1 classification performance. Data processing steps include the following. SNV genotypes are encoded based on alternative allele dosage (0, 1, 2). To ensure statistical robustness, only SNVs with a minor allele frequency > 0.05 are considered. Population structure is accounted for with principal component analysis (PCA) of the genotype matrix, with top components included as covariates. Health & Exposure survey responses are filtered, and variables with more than 5% missing responses and more than 95% single response frequency are removed. Variables with strong collinearity or sex-dependent responses are also removed. Numerical and ordered categorical variables are used as-is; binary variables were encoded as 0/1; unordered categorical and free-text variables were excluded. The marginal effect a of a given SNV ( G ) on trait ( Y ) is first quantified as: where a 0 is the intercept and is the error term. To estimate the genetic main effect b 1 , in the presence of the environment ( E ), the model can be extended to: where b 2 represents the environmental main effect, and b 3 the interaction effect. Covariates from PCA and other relevant confounders such as sex are included in the models to account for population structure and confounding. To detect significant GxE interactions, the linear models described above can be fit to all selected G,E pairs. The relationship between the genetic marginal effects ( a 1 ) can then be modeled in the absence of E and the genetic main effects ( b 1 ) in the presence of E with linear regression. As previously demonstrated 17 , in the absence of a true GxE interaction effect, b 1 and a 1 are linearly correlated. This developed hypothesis leverages this property to systematically test for the presence of GxE interactions in the data. Potentially interacting G,E pairs can be ranked based on their deviation from this expected correlation and reviewed for biological interpretation. Discussion The PEGS DREAM Challenge demonstrates the power of crowdsourced problem-solving in leveraging multidimensional data to investigate the etiology of a complex condition—hypercholesterolemia. This initiative brought together diverse expertise from across the world to develop integrative models and testable hypotheses, yielding both methodological innovation and new biological insights. The models presented in the PEGS DREAM Challenge demonstrate that integrative approaches combining self-reported exposome and health survey-based data with genomic information can outperform models based on PGS alone. This finding underscores the value of incorporating multi-dimensional data, such as those in the PEGS cohort, to capture the complex interplay of genetic and environmental factors in diseases like hypercholesterolemia. This Challenge exemplifies how community engagement in a competitive yet collaborative setting can accelerate methodological advances and generate new research directions in precision environmental health. In Challenge Task 1, the top-performing teams substantially outperformed the benchmark PGS (AUROC = 0.7358), with AUROCs of 0.7933 and 0.7919, respectively, highlighting the usefulness of integrating survey-based health and exposure data with genomic information. Team Spider Bobs relied on correlation networks derived from ExWAS results to identify informative features, while Team Nonsense-Mediated Decay combined ensemble learning with PGS to build a hybrid model. The comparable performance of these two methodologically distinct approaches demonstrates that robust prediction of hypercholesterolemia status can be achieved using flexible combinations of multimodal data. Importantly, these models show that self-reported health and exposure variables—often underutilized in clinical prediction—add measurable value to disease classification beyond genetic risk scores. Challenge Task 2 extended the utility of the PEGS dataset beyond classification, asking participants to generate novel hypotheses regarding hypercholesterolemia risk. These ideation submissions revealed potential biological mechanisms and environmental contributors that warrant further investigation. For example, Team Spider Bobs identified differentially methylated probes in genes such as ELOVL2 and FHL2 , implicating lipid metabolism and cardiovascular regulation pathways. Their finding of a significant association between HLA-DPB1 alleles and hypercholesterolemia echoes known links between HLA variation and lipid phenotypes, suggesting immune-mediated contributions to metabolic disease risk. Unexpected associations of hypercholesterolemia with self-reported exposure to dyes and diagnosis with German Measles also emerged. Although these findings may reflect latent confounding (e.g., socioeconomic status), they illustrate how unconventional variables can surface through hypothesis-free exploration and generate new research questions. Team Nonsense-Mediated Decay, meanwhile, proposed a systematic method for detecting GxE interactions using deviations from expected genotype-phenotype correlations in the presence of environmental variables. Their approach, rooted in modeling marginal versus conditional genetic effects, highlights the utility of theoretical expectations to identify non-additive interactions. This strategy is broadly applicable to other traits and datasets, offering a reproducible framework for large-scale GxE analysis. Beyond the findings themselves, the Challenge showcased key methodological innovations. The use of synthetic data allowed teams to develop and refine models while maintaining participant privacy. The provision of resources such as PEGS correlation globes and ExWAS results enabled participants to perform informed feature selection and contextual interpretation. These tools, combined with Docker-based model evaluation, ensured reproducibility and transparency in model scoring, setting a standard for future data science competitions involving sensitive health data. Conclusions Ultimately, the PEGS DREAM Challenge fostered a fertile environment for developing new approaches to integrating genomic, environmental, and social determinants of health data. The top-performing models offered not only improved classification tools but also a roadmap for incorporating diverse data types into disease modeling. These models could be clinically relevant and help build robust and translatable predictive tools for personalized medicine with validation against objective clinical outcomes using data such as electronic health records. The ideation task reinforced the importance of hypothesis generation in complex trait research, particularly when guided by novel, interpretable, and potentially actionable variables. Taken together, the results of the Challenge suggest that integrative, crowdsourced approaches can reveal both methodological and biological insights into diseases shaped by GxE, which will be critical as biobank-scale datasets continue to expand. Methods PEGS DREAM Challenge Data Personalized Environment and Genes Study (PEGS) cohort Originally established in 2002 as the Environmental Polymorphisms Registry (EPR) to recruit participants for ongoing research at NIEHS by convenience sampling at community events, PEGS was renamed in 2022. This racially and ethnically diverse North Carolina-based cohort is a repository of data on medications, health outcomes, environmental exposures, lifestyle factors, genomic data, and geospatial estimates of exposure (Figure 1). PEGS participants complete three surveys—the Health & Exposure Survey and the Internal and External Exposome Surveys—and provide biological samples. Participants can consent to be called back for additional tissue collection for add-on studies. Further details of PEGS can be found at https://www.niehs.nih.gov/research/atniehs/labs/crb/studies/pegs/index.cfm. In contrast to typical studies that focus on a single disease or environmental exposure, PEGS gathers data on a wide array of diseases and environmental exposures, including dietary and lifestyle factors, in conjunction with genomic data. The overarching aim of PEGS is to empower researchers to unravel the etiology of disease and uncover how environmental, dietary, lifestyle, and genetic factors collectively impact human health. The PEGS cohort (N=19,445) comprises approximately two-thirds female (62.3%; 12,120) and one-third male (37.6%; 7,298) participants ranging in age from 18.4 to 98.3 years, with a mean age of 50.2 years (at completion of the Health & Exposure Survey). The self-reported racial makeup of the cohort is two-thirds White (63%; 12,279), slightly over one-quarter Black (27.6%; 5,361), and 4.5% who identify as another race (868). Additionally, 5.0% self-reported their ethnicity as Hispanic (979). This diverse cohort includes participants of varying socioeconomic status and education levels, enabling research on disease risk in multiple populations. Further, this diversity supports broadly applicable results and can help uncover health disparities that occur due to disproportionate environmental exposures for certain populations. Data provided to PEGS DREAM Challenge teams PEGS DREAM Challenge teams were provided access to data from PEGS Data Freeze 3.1 for PEGS participants who completed the Health & Exposure Survey (N = 9,449) (see Table 1). The questionnaire-based data include information on demographics, health, environmental exposures, socioeconomic status, and lifestyle factors (Tables 1 and 2). Genomic data on SNVs and indel genotypes, structural variant calls, HLA genotypes, estimated aggregate telomeric content, and ancestry estimations were derived from WGS. Genome-wide methylation profiling data were obtained using the Infinium MethylationEPIC v1.0 BeadChip Kit (Figure 6). Geospatial data include exposure estimates and proximity to hazards calculated using geospatial linkages with various databases, linkages from the MERRA-2 project providing estimates of climate and environmental metrics from satellite observations, and SVI and EJI data (Table 3). Table 1. PEGS data components. The various data components available in the PEGS cohort. Data reported from PEGS Data Freeze 3.1 created on 6/27/2023. Category Component Description Survey Data Demographic and administrative data Demographics, consent, addresses, and administrative data for all participants Health & Exposure Survey Demographics, health, family history of disease, environmental exposures, socioeconomic status, and lifestyle factors External Exposome Survey (Exposome A) Residential and occupational environmental exposures Internal Exposome Survey (Exposome B) Medication use, physical activity, stress, sleep, diet, genetics, and reproductive history Geospatial Data Hazards data Exposure estimates and proximity measures calculated using geospatial linkages from various databases MERRA-2 Data (Earthdata) Geospatial data linkages from the MERRA-2 project containing consistent estimates of climate and environmental metrics from a range of satellite-based environmental observations Social Vulnerability Index (SVI) data Geospatial data linkages for the CDC/ATSDR SVI containing summaries of social determinants of health at the census-tract level Environmental Justice Index (EJI) data Geospatial data linkage for the CDC/ATSDR EJI containing summaries of environmental, social, and health factors at the census-tract level Genomic data Single nucleotide variants (SNVs) SNV and small indel genotypes derived from the whole genome sequencing (WGS) data in PLINK’s .bed/.bim/.fam format Structural variants Structural variant calls generated from the WGS data in .vcf format consisting of large deletions, duplications, and inversions Human leukocyte antigens (HLA) Genotypes HLA genotypes identified from the WGS data for 20 HLA genes with up to six digits of specificity Telomeric content Aggregate telomeric content estimated from WGS reads reported as telomeric reads per GC content-matched million reads Local and global ancestry estimations Inferred local ancestry per chromosome after haplotype phasing and global estimates of percent ancestry for each participant Methylation data Genome-wide methylation profiling data using the Infinium MethylationEPIC v1.0 BeadChip Kit targeting 866,297 CpG sites Table 2. Categories of questions in the PEGS Health & Exposure Survey, External Exposome Survey, and Internal Exposome Survey. The table shows the high-level survey question categories of the surveys administered to PEGS participants. Data reported from PEGS Data Freeze 3.1 created on 6/27/2023. Health & Exposure Survey About Your Family’s Health Diabetes and Endocrine Neurologic About Your General Health Digestive Occupation About Your Home Life Exposures Renal About Your Mood Fatigue Reproductive (Females Only) Bones, Joints, and Muscles Hematological Reproductive (Males Only) Cancer Immune Respiratory Cardiovascular Lifestyle Skin, Eyes, and Hair External Exposome (Exposome A) Characteristics of Current and Past Residences: • Agricultural Property Use • Garage and Basement • Heating and Cooling • Pesticides and Insecticides • Pets • Surrounding Area • Walls and Flooring • Water and Dampness Chemical and Metal Exposures at Work Ultraviolet Light Exposure Workplace Characteristics Hobby Exposures Internal Exposome (Exposome B) Chemotherapy/Radiation Therapy Physical Activity Dietary Behavior Reproductive History (Females Only) Dietary Intake Sleep Genetic History Stress Infectious Disease Vitamins, Minerals, and Other Supplement Use Medications Twin/Triplet Siblings and Birth Order Other Table 3. Summary of PEGS geospatial data. An overview of the geocoding and GIS data linkages available in the PEGS cohort. Data reported from PEGS Data Freeze 3.1 created on 6/27/2023. Source Description Examples Geocodes (GIS) Geocoded data from multiple participant-provided addresses at the time of initial enrollment, completion of the Health & Exposure Survey, completion of the External Exposome Survey, the longest-lived childhood address, and the longest-lived adulthood address from the External Exposome Survey Geographic coordinates (latitude and longitude) from multiple participant-provided addresses Hazards Exposure estimates computed from Department of Transportation (DOT) data Information on train tracks, rail depots, and roadways, such as total major roadway length and distance to the nearest rail depot Hazards Exposure estimates computed from Federal Aviation Administration (FAA) data Information on aircraft departure and arrival sites (e.g., distance to the nearest airport) Hazards Exposure estimates computed from Federal Communications Commission (FCC) data Information on cellular network towers (e.g., nearest cell tower) Hazards Exposure estimates computed from the North Carolina Department of Environmental Quality (NCDEQ) Distance to multi-pollutant point sources such as swine caged feeding operations (CAFOs), hazardous waste sites, hazardous spill sites, EPA superfund sites, and wastewater treatment plant release sites Hazards Exposure estimates computed from Nuclear Regulatory Commission (NRC) data Distance to nuclear power stations Hazards Exposure estimates computed from Atmospheric Composition and Analysis Group (ACAG) data Particulate matter concentrations such as PM2.5 total, PM2.5 sulfate, and PM2.5 black carbon and other Hazards Exposure estimates computed from Center for Air, Climate, and Energy Solutions (CACES) data Concentrations for multiple pollutants such as carbon monoxide, nitrogen dioxide, and ozone concentration Hazards Exposure estimates computed from Toxics Release Inventory (TRI) data Emissions for chemicals of interest such as benzene, ethylbenzene, xylene, and toluene MERRA-2 data (Earthdata) Geospatial data linkages from the Modern Era Retrospective Analysis for Research and Applications (MERRA-2) project to assimilate a range of satellite-based environmental observations into a consistent estimate of climate and environmental metrics Particulate, gas, meteorological, and health-relevant exposure indicators such as dust sedimentation, organic carbon emission bin, SO 2 biomass burning emissions, and sea-level pressure Social Vulnerability Index (SVI) Geospatial data linkages for CDC/ATSDR SVI designed to consistently quantify multiple social determinants of health across the United States over time Summaries of social determinants of health at the census-tract level, including an overall index, four component indices (socioeconomic status, household characteristics, racial and ethnic minority status, and housing type/transportation), and source variables used to compute each index component (e.g., poverty, education, overcrowding, access to a vehicle) Environmental Justice Index (EJI) Geospatial data linkages for CDC/ATSDR EJI containing summaries and ranks of the cumulative impacts of environmental injustice on health at the census-tract level Ranks for each census tract based on 36 environmental, social, and health factors grouped into 10 domains and three overarching modules: environmental burden, social vulnerability, and health vulnerability The dataset (N = 9,449) was filtered to remove outliers, related individuals determined from WGS data, and individuals with mismatched self-reported and WGS-inferred race (N = 9,184). The filtered data were split into individuals with and without WGS data. Each subset was then randomly split into thirds to create training (n = 3,062), validation (n = 3,062), and test (n = 3,060) subsets. Accordingly, the training, validation, and test subsets each consisted of approximately 1,500 individuals with and 1,500 individuals without sequencing data. The test subset was not shared with Challenge participants and was held out to score Final round submissions. Table 4 provides the demographics of participants included in the DREAM Challenge data. Table 4. PEGS DREAM Challenge participant demographics and data availability. Demographics of participants and survey and WGS data availability in the PEGS DREAM Challenge data in the training, validation, and test subsets. Age was computed at the time of completion of the Health & Exposure Survey. The PEGS DREAM Challenge data was created from PEGS Data Freeze 3.1 created on 6/27/2023. Demographic variable Training data n (%) Validation data n (%) Test data n (%) Total N 3062 (33.3) 3062 (33.3) 3060 (33.3) Sex Male 994 (32.5) 993 (32.4) 1039 (34) Female 2068 (67.5) 2069 (67.6) 2021 (66) Race Other 133 (4.3) 131 (4.3) 140 (4.6) Black or African American 696 (22.7) 678 (22.1) 620 (20.3) White 2157 (70.4) 2179 (71.2) 2214 (72.4) NA 76 (2.5) 74 (2.4) 86 (2.8) Ethnicity Non-Hispanic/Non-Latino 2886 (94.3) 2887 (94.3) 2905 (94.9) Hispanic/Latino 119 (3.9) 130 (4.2) 112 (3.7) NA 57 (1.9) 45 (1.5) 43 (1.4) Education 12th grade or less 528 (17.2) 493 (16.1) 501 (16.4) College, technical or vocational 887 (29) 915 (29.9) 912 (29.8) Bachelor's degree 828 (27) 811 (26.5) 838 (27.4) Graduate or professional degree 800 (26.1) 819 (26.7) 783 (25.6) NA 19 (0.6) 24 (0.8) 26 (0.8) Income Less than $20,000 432 (14.1) 400 (13.1) 417 (13.6) $20,000 to 49,999 896 (29.3) 937 (30.6) 907 (29.6) $50,000 to 79,999 740 (24.2) 728 (23.8) 718 (23.5) $80,000 or more 892 (29.1) 889 (29) 906 (29.6) NA 102 (3.3) 108 (3.5) 112 (3.7) Age in years (SD) 50 (16) 50.4 (15.9) 50.2 (16) Health & Exposure Survey completed N= 3062 (100) 3062 (100) 3060 (100) External Exposome completed No 2001 (65.3) 1992 (65.1) 2018 (65.9) Yes 1061 (34.7) 1070 (34.9) 1042 (34.1) Internal Exposome completed No 2156 (70.4) 2151 (70.2) 2155 (70.4) Yes 906 (29.6) 911 (29.8) 905 (29.6) WGS available No 1547 (50.5) 1547 (50.5) 1546 (50.5) Yes 1515 (49.5) 1515 (49.5) 1514 (49.5) Hypercholesterolemia status Yes 1018 (33.2) 1011 (33.0) 983 (32.1) No 2012 (65.7) 2001 (65.3) 2029 (66.3) In the test dataset, 983 (32.8%) PEGS participants self-reported a diagnosis of hypercholesterolemia in the Health & Exposure Survey. Table 4 shows the number of PEGS participants available for the specific PEGS data component. In the PEGS DREAM Challenge data, WGS data were available for 4,544 participants, and Internal and External Exposome Survey data were available for 2,722 and 3,173 participants, respectively. Synthetic data The synthpop v1.8 library in R was used to create synthetic data from the original tabular PEGS data. For each data file stratified into training/validation/testing sets, a synthetic version equal in size was created using a random forest model, sequentially for each feature. By generating synthetic data for each table separately, inter-table correlation was not preserved. However, the data were appropriate for developing methods that were validated during this Challenge phase. For large survey and genomic data tables, independent resampling and error injection were performed to match the data types in the original dataset but remove correlation between features. The synthetic dataset contained training (_train.*) and validation (*_val.*) data that reflect the structure and sample sizes of the original PEGS dataset. In the synthetic dataset, “NA” was used for unmeasured missing phenotype variables to preserve the original data structure. The synthetic and original datasets contain the same number of individuals in each file. Multiple measures were taken to protect the privacy of PEGS participants. First, all data were de-identified by removing all personally identifiable information (PII) and protected health information (PHI). Second, the data were anonymized by replacing participant IDs with generated IDs for each data component. Third, Challenge participants were not provided access to the original PEGS data. Only the synthetic data were available for download for model construction and code development. The submitted models were evaluated using the original PEGS data that were unavailable for download by Challenge participants. Challenge Details The PEGS DREAM Challenge invited teams to develop models to identify the myriad factors affecting hypercholesterolemia and improve understanding of the etiology of this complex disease. Determining modifiable factors can help reduce the likelihood of developing hypercholesterolemia and its associated health complications. The Challenge was launched on May 1, 2024, and a Leaderboard round ran from May 10 to August 6, 2024. The Final round ran from August 6 to August 19, 2024. Winners were announced on August 31, 2024 and presented their models and findings at the Regulatory & Systems Genomics with DREAM Challenges Conference (RSGDREAM2024) in Madison, Wisconsin, USA (https://www.iscb.org/rsgdream2024/home). Challenge teams could take part in one or both distinct tasks. The goal of Task 1was to classify individual hypercholesterolemia disease status in the held-out test dataset using combinations of the health, exposomic, geospatial, and genomic data available for the PEGS cohort. The goal of Task 2, an ideation challenge, was to develop novel hypotheses and models using the multi-dimensional PEGS data to improve understanding of the etiology of hypercholesterolemia beyond conventional clinical and genetic risk factors. To ensure rigorous evaluation while safeguarding participant privacy, the Challenge was implemented using a model-to-data framework 18 . In this paradigm, the PEGS cohort data were stored securely on the Challenge platform, and participants submitted containerized models that were executed against hidden validation datasets. This approach, successfully pioneered in previous DREAM Challenges, enabled participants to develop innovative methods without direct access to confidential health and genomic data while organizers ensured unbiased evaluation. This framework was used in previous Challenges that include the Digital Mammography DREAM Challenge (2017), Patient Mortality Prediction EHR DREAM Challenge (2019), COVID-19 EHR DREAM Challenge (2020), and CD2H NLP Sandbox (2021). Challenge Task 1 – Disease Classification Figure 4 shows the workflow for Task 1 of the PEGS DREAM Challenge. Participants were provided with synthetic data with the same format and structure as the original PEGS data. For details, see the synthetic data section. The training and validation subsets were available for model training and model optimization, respectively, and participants could combine or split the subsets in various proportions. The test dataset was not shared with challenge contributors and was used for scoring and evaluation. Participants were asked to integrate the multi-dimensional components of PEGS, comprising health, exposure, genomic, and geospatial data, to create a model that surpassed the classification accuracy of an AUROC = 0.7358. The AUROC value was obtained from a model using a hypercholesterolemia PGS computed for PEGS participants using the weights and metadata from the hypercholesterolemia PGS by Weissbrod et al. 6 from the Polygenic Score Catalog 7,8 as the main predictor. This hypercholesterolemia PGS was chosen because it included the highest number of genomic variants in the score compared to other hypercholesterolemia scores in the catalog and included individuals from multiple ancestries in the evaluation set, mirroring the diversity of the PEGS cohort. A detailed description of PGS computation for PEGS participants is provided in Schaid et al. 19 Teams uploaded Docker containers and writeups in .docx or .pdf format to their Synapse Project workspace. The submitted Docker container produced a single output file classifying hypercholesterolemia status for the held-out test dataset in a two-column .csv file containing participant IDs and disease probability. The submitted Docker containers were run on the original PEGS data, with a limit of five model submissions per day in the Leaderboard round and one model submission per day in the Final round. Scoring was based on the AUROC of each submitted model in the validation subset for the Leaderboard round and in the test dataset for the Final round. AUPRC was employed as a tie-breaking metric. Challenge Task 2 – Ideation Challenge In Task 2, Challenge participants leveraged the comprehensive PEGS dataset to develop bold and novel hypotheses and models, integrating its health, exposure, geospatial, and genomic components to provide multi-dimensional insights. This challenge was aimed at harnessing collective creativity to generate a pipeline of fresh ideas to address complex challenges. Participants proposed novel research questions and methodologies, with the goals of fostering ideation related to emerging omics technologies and their applications, incentivizing outside-the-box thinking beyond established analytical approaches, and promoting collaboration and interdisciplinary approaches for holistic exploration. Teams submitted a writeup in .docx or .pdf format documenting their hypotheses and models, which were scored by a panel of judges based on creativity, significance, interpretability, innovation, utility, feasibility, and potential translational impact. Submissions with working or detailed models or specific mathematical model descriptions were given higher scores. Participants were encouraged to submit models involving GxE interactions, genome-wide environment interaction studies (GWEIS), and similar methods. Innovation was emphasized to encourage participants to explore novel approaches. Challenge Resources The PEGS DREAM Challenge website (https://www.synapse.org/PEGS) describes in detail the PEGS cohort, data available to participants, and two Challenge tasks and provides submission guidelines and evaluation criteria. The website also outlines the Challenge timeline, participation criteria, and conditions for data use. Additionally, the AUROCs of the models submitted for Task 1 computed in the validation data were published live on the Challenge website during the Leaderboard round to provide participants with feedback and enable them to improve their models for the Final round. In addition to the Challenge website, participants had access to several other PEGS resources for data exploration, hypothesis generation, and investigation of results of prior analyses in the PEGS data. The PEGS Explorer (https://pegsexplorer.niehs.nih.gov/) web application shares published results of ExWAS conducted in PEGS data and visualizations of the complex correlations among the exposures through correlation globes 10,11 . Participants could explore and use these results for the Challenge. Participants could also request access to the i2b2 self-service web-based tool for data exploration, which enabled participants to explore de-identified and aggregated PEGS data by building queries. The results assisted with data exploration for model construction and hypothesis generation. Participants were also provided with access to code libraries for the PEGS data as a reference for development efforts. This included a library containing common utilitarian functions for ingesting and analyzing the PEGS data (https://github.com/fsakhtari/PEGS_common/blob/master/pegs_common_utils.R) and example scripts (https://github.com/nathanielmacnell/PEGStools) for working with the PEGS geospatial data. All methods were carried out in accordance with relevant guidelines and regulations. The PEGS study protocol and/or the use of PEGS data for this study were approved by the NIEHS IRB, protocol 04-E-0053. Informed consent was obtained from all participants and/or their legal guardians. Declarations Ethics approval and consent to participate Not applicable Consent for publication Not applicable Competing interests The authors declare that they have no competing interests. Funding This work was funded in part by the intramural research program of the National Institute of Environmental Health Sciences. The contributions of the NIH authors were made as part of their official duties as NIH federal employees, are in compliance with agency policy requirements, and are considered Works of the United States Government. However, the findings and conclusions presented in this paper are those of the author(s) and do not necessarily reflect the views of the NIH or the U.S. Department of Health and Human Services. Authors’ contributions F.S.A., The PEGS DREAM Challenge Community, G.S.A, and A.A.M-R. conceptualized the study. F.S.A., J.F., J.J., V.V., E.C., A.S-Y., M-T.H., A.L., F.M., G.A., S.M., V.C., J.A.,D.C.F., C.P.S., and J.E.H curated the data used in the study and performed the formal analysis. F.S.A. and A.A.M-R. provided project administration and wrote the original manuscript draft. G.S.G. supervised the study. F.S.A. and G.S.G. validated the study results. J.F., J.J., V.V., E.C., A.S-Y., M-T.H., A.L., F.M., G.A., S.M., V.C., J.A., D.C.F., C.P.S., and J.E.H reviewed and edited the manuscript. A.A.M-R. acquired funding for the study and was involved in investigation and methodology. All authors approved the submitted version of the manuscript. Acknowledgements We would like to thank the PEGS participants for their contributions to this work. We would also like to express our sincere appreciation to Sharon Soucek in the Office of Technology Transfer at NIEHS for support and expertise regarding the data use agreements that enable data sharing and collaborative research projects with PEGS. We would like to thank Sage Bionetworks for hosting the PEGS DREAM Challenge on their platform and providing the necessary infrastructure support. We also thank Hannah Collins Cakar for manuscript support. Data Availability The PEGS dataset analyzed in this study is not publicly available because it contains sensitive human participant data and is subject to ethical and privacy restrictions. De-identified data supporting the findings of this study are available from the corresponding author upon reasonable request, subject to review and approval by the NIEHS IRB, compliance with applicable ethical and legal requirements, and execution of any required data use agreement. Requests for access should be directed to [email protected] or through a web form inquiry: https://www.niehs.nih.gov/research/atniehs/labs/crb/studies/pegs/collaboration/proposal. PEGS DREAM Challenge participants were required to provide code as Docker containers to run their models for evaluation. Code libraries for working with PEGS data were provided as resources (e.g., https://github.com/fsakhtari/PEGS_common/blob/master/pegs_common_utils.R, https://github.com/nathanielmacnell/PEGStools). References Wong, N. D., Lopez, V., Tang, S. & Williams, G. R. Prevalence, treatment, and control of combined hypertension and hypercholesterolemia in the United States. Am J Cardiol 98 , 204–208 (2006). https://doi.org/https://doi.org/10.1016/j.amjcard.2006.01.079 Ordovas, J. M. et al. Gene-diet interaction in determining plasma lipid response to dietary intervention. Atherosclerosis 118 Suppl , S11–27 (1995). Ordovas, J. M. & Shen, J. Gene–environment interactions and susceptibility to metabolic syndrome and other chronic diseases. J Periodontol 79 , 1508–1513 (2008). Ye, S. Q. & Kwiterovich, P. O., Jr. Influence of genetic polymorphisms on responsiveness to dietary fat and cholesterol. Am J Clin Nutr 72 , 1275s–1284s (2000). https://doi.org/10.1093/ajcn/72.5.1275s Meyer, P. & Saez-Rodriguez, J. Advances in systems biology modeling: 10 years of crowdsourcing DREAM challenges. Cell Syst 12 , 636–653 (2021). https://doi.org/10.1016/j.cels.2021.05.015 Weissbrod, O. et al. Leveraging fine-mapping and multipopulation training data to improve cross-population polygenic risk scores. Nat Genet 54 , 450–458 (2022). Lambert, S. A. et al. Enhancing the Polygenic Score Catalog with tools for score calculation and ancestry normalization. Nat Genet 56 , 1989–1994 (2024). https://doi.org/10.1038/s41588-024-01937-x Lambert, S. A. et al. The Polygenic Score Catalog as an open database for reproducibility and systematic evaluation. Nat Genet 53 , 420–425 (2021). https://doi.org/10.1038/s41588-021-00783-5 Pedregosa, F. et al. Scikit-learn: Machine learning in Python. J Mach Learn Res 12 , 2825–2830 (2011). Lloyd, D. et al. Questionnaire-based exposome-wide association studies for common diseases in the Personalized Environment and Genes Study. Exposome 4 , osae002 (2024). Lloyd, D. et al. Interactive data sharing for multiple questionnaire-based exposome-wide association studies and exposome correlations in the Personalized Environment and Genes Study. Exposome 4 , osae003 (2024). Ritchie, M. E. et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 43 , e47–e47 (2015). Chang, C. C. et al. Second-generation PLINK: rising to the challenge of larger and richer datasets. Gigascience 4 , s13742–13015–10047–13748 (2015). Kennedy, R. B., Ovsyannikova, I. G., Vierkant, R. A., Jacobson, R. M. & Poland, G. A. Effect of human leukocyte antigen homozygosity on rubella vaccine–induced humoral and cell-mediated immune responses. Hum Immun 71 , 128–135 (2010). Lambert, N. D. et al. Polymorphisms in HLA-DPB1 are associated with differences in rubella virus–specific humoral immunity after vaccination. J Infect Dis 211 , 898–905 (2015). Hübschen, J. et al. Challenges of measles and rubella laboratory diagnostic in the era of elimination. Clin Microbiol Infect 23 , 511–515 (2017). Zhu, X. et al. An approach to identify gene-environment interactions and reveal new biological insight in complex traits. Nat Commun 15 , 3385 (2024). https://doi.org/10.1038/s41467-024-47806-3 Guinney, J. & Saez-Rodriguez, J. Alternative models for sharing confidential biomedical data. Nat Biotechnol 36 , 391–392 (2018). https://doi.org/10.1038/nbt.4128 Schaid, D. J. et al. Polygenic scores and social determinants of health: Their correlations and potential biases. Hum Genet Genom Adv 6 (2025). https://doi.org/10.1016/j.xhgg.2024.100389 Additional Declarations No competing interests reported. Supplementary Files PEGSDREAMChallengemanuscriptsupplement26092025.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 18 May, 2026 Reviewers agreed at journal 16 Apr, 2026 Reviewers invited by journal 13 Apr, 2026 Editor assigned by journal 13 Apr, 2026 Editor invited by journal 13 Apr, 2026 Submission checks completed at journal 06 Apr, 2026 First submitted to journal 06 Apr, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9254914","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":624041595,"identity":"2d8766b7-cb0d-4542-b61c-555a7ae821ab","order_by":0,"name":"Farida S. Akhtari","email":"","orcid":"","institution":"National Institute of Environmental Health Sciences","correspondingAuthor":false,"prefix":"","firstName":"Farida","middleName":"S.","lastName":"Akhtari","suffix":""},{"id":624041596,"identity":"49bc8492-c9de-4fde-abd2-74bc2d62b73e","order_by":1,"name":"Johannes Falk","email":"","orcid":"","institution":"Constructor University","correspondingAuthor":false,"prefix":"","firstName":"Johannes","middleName":"","lastName":"Falk","suffix":""},{"id":624041597,"identity":"9581b923-8fc8-413a-8417-428956444606","order_by":2,"name":"Jyoti Jyoti","email":"","orcid":"","institution":"Constructor University","correspondingAuthor":false,"prefix":"","firstName":"Jyoti","middleName":"","lastName":"Jyoti","suffix":""},{"id":624041598,"identity":"3430c48c-8e24-4ada-b7f1-aacaf6b5199b","order_by":3,"name":"Venetia Voutsa","email":"","orcid":"","institution":"Constructor University","correspondingAuthor":false,"prefix":"","firstName":"Venetia","middleName":"","lastName":"Voutsa","suffix":""},{"id":624041599,"identity":"535b0748-6d19-4004-a643-40a2637cface","order_by":4,"name":"Eda Cakir","email":"","orcid":"","institution":"Constructor University","correspondingAuthor":false,"prefix":"","firstName":"Eda","middleName":"","lastName":"Cakir","suffix":""},{"id":624041600,"identity":"f23d6cfd-5f76-4d31-80c8-058968eeb8f4","order_by":5,"name":"Ali Salehzadeh-Yazdi","email":"","orcid":"","institution":"Constructor University","correspondingAuthor":false,"prefix":"","firstName":"Ali","middleName":"","lastName":"Salehzadeh-Yazdi","suffix":""},{"id":624041601,"identity":"8ebd4d7b-716d-4711-8dc6-75797cd52af8","order_by":6,"name":"Marc-Thorsten Hütt","email":"","orcid":"","institution":"Constructor University","correspondingAuthor":false,"prefix":"","firstName":"Marc-Thorsten","middleName":"","lastName":"Hütt","suffix":""},{"id":624041602,"identity":"5abf6300-f4fb-4cfb-a61e-625b205318db","order_by":7,"name":"Alessandro Lussana","email":"","orcid":"","institution":"European Bioinformatics Institute","correspondingAuthor":false,"prefix":"","firstName":"Alessandro","middleName":"","lastName":"Lussana","suffix":""},{"id":624041603,"identity":"a3226dfe-adfb-4423-9078-9fdc2ef8a272","order_by":8,"name":"Federico Marotta","email":"","orcid":"","institution":"European Molecular Biology Laboratory","correspondingAuthor":false,"prefix":"","firstName":"Federico","middleName":"","lastName":"Marotta","suffix":""},{"id":624041604,"identity":"cd7bafd1-dd85-4922-8702-2f2c668e1dbd","order_by":9,"name":"The PEGS DREAM Challenge Community","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"The","middleName":"PEGS DREAM Challenge","lastName":"Community","suffix":""},{"id":624041605,"identity":"2d797c50-ee97-4454-ade6-0d0bf56eda1b","order_by":10,"name":"Geoffrey S. Ginsburg","email":"","orcid":"","institution":"National Institutes of Health","correspondingAuthor":false,"prefix":"","firstName":"Geoffrey","middleName":"S.","lastName":"Ginsburg","suffix":""},{"id":624041606,"identity":"728a16dd-6e16-4daf-9e57-819df4e8608f","order_by":11,"name":"Gaia Andreoletti","email":"","orcid":"","institution":"Sage Bionetworks","correspondingAuthor":false,"prefix":"","firstName":"Gaia","middleName":"","lastName":"Andreoletti","suffix":""},{"id":624041607,"identity":"3fe0e1be-e41a-4286-b5af-8dac0f36477c","order_by":12,"name":"Serghei Mangul","email":"","orcid":"","institution":"Sage Bionetworks","correspondingAuthor":false,"prefix":"","firstName":"Serghei","middleName":"","lastName":"Mangul","suffix":""},{"id":624041608,"identity":"544fe34b-46e7-4266-9fbc-02649a3a0a12","order_by":13,"name":"Verena Chung","email":"","orcid":"","institution":"Sage Bionetworks","correspondingAuthor":false,"prefix":"","firstName":"Verena","middleName":"","lastName":"Chung","suffix":""},{"id":624041609,"identity":"41d71a3e-202f-403b-af6b-41159c2f54ce","order_by":14,"name":"Jacob Albrecht","email":"","orcid":"","institution":"Sage Bionetworks","correspondingAuthor":false,"prefix":"","firstName":"Jacob","middleName":"","lastName":"Albrecht","suffix":""},{"id":624041610,"identity":"9a0dd08d-9cc9-4947-ad04-9a41212461c5","order_by":15,"name":"David C. Fargo","email":"","orcid":"","institution":"National Institute of Environmental Health Sciences","correspondingAuthor":false,"prefix":"","firstName":"David","middleName":"C.","lastName":"Fargo","suffix":""},{"id":624041611,"identity":"c36d0109-3925-4c97-9075-53ceba3530d5","order_by":16,"name":"Charles P. Schmitt","email":"","orcid":"","institution":"National Institute of Environmental Health Sciences","correspondingAuthor":false,"prefix":"","firstName":"Charles","middleName":"P.","lastName":"Schmitt","suffix":""},{"id":624041612,"identity":"237c78dc-54d4-4ec1-9170-27bb2cd105a8","order_by":17,"name":"Janet E. Hall","email":"","orcid":"","institution":"National Institute of Environmental Health Sciences","correspondingAuthor":false,"prefix":"","firstName":"Janet","middleName":"E.","lastName":"Hall","suffix":""},{"id":624041613,"identity":"f99e5dbb-88ea-4907-9655-465987c04157","order_by":18,"name":"Alison A. Motsinger-Reif","email":"data:image/png;base64,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","orcid":"","institution":"National Institute of Environmental Health Sciences","correspondingAuthor":true,"prefix":"","firstName":"Alison","middleName":"A.","lastName":"Motsinger-Reif","suffix":""}],"badges":[],"createdAt":"2026-03-28 19:23:44","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9254914/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9254914/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107614285,"identity":"7a3f9957-4223-4721-bf3a-d8da0af9b00c","added_by":"auto","created_at":"2026-04-23 09:07:56","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":836008,"visible":true,"origin":"","legend":"\u003cp\u003eOverview of available PEGS data. Data reported from PEGS Data Freeze 3.1 created on 6/27/2023.\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9254914/v1/7248598650f08b97510a320f.jpeg"},{"id":107705850,"identity":"002e9ee6-06b0-4692-ae76-33ce34633514","added_by":"auto","created_at":"2026-04-24 09:15:32","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":382146,"visible":true,"origin":"","legend":"\u003cp\u003eSummary of the genomic data available in the PEGS cohort. Whole-genome sequencing was used to obtain single nucleotide variants consisting of common and rare variants, structural variant calls, human leukocyte antigen (HLA) genotypes for 20 HLA genes, aggregate telomeric content estimation, inferred local ancestry per chromosome, and global estimates of percent ancestry. Genome-wide methylation profiling data were obtained using the Infinium MethylationEPIC v1.0 BeadChip Kit. Data reported from PEGS Data Freeze 3.1 created on 6/27/2023.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9254914/v1/2c0cb3556b803b12e4bff068.png"},{"id":107614286,"identity":"c59778dd-7273-48fc-9039-1841c8bf342e","added_by":"auto","created_at":"2026-04-23 09:07:56","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":327274,"visible":true,"origin":"","legend":"\u003cp\u003ePEGS GIS data. Overview of the multi-pollutant point sources used to estimate exposure indicators for PEGS participants using their geocoded addresses. Data reported from PEGS Data Freeze 3.1 created on 6/27/2023.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-9254914/v1/c32968c46c3b2d6a2f9b5c65.png"},{"id":108006014,"identity":"628709ab-e8f2-4d3a-bd29-0b694a94de42","added_by":"auto","created_at":"2026-04-28 12:51:54","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":141749,"visible":true,"origin":"","legend":"\u003cp\u003ePEGS DREAM Challenge workflow for Task 1. The multi-dimensional PEGS DREAM Challenge data (phenotype, genomic, and environmental data) were randomly and equally split into training, validation, and test data subsets. Synthetic data were available for Challenge participants to download for model construction. Challenge data were split into training (N=3062) and validation (N=3062) subsets. Participants could submit up to five model submissions per day, packaged as Docker containers. The final submitted models were evaluated against a held-out test dataset (N=3060), and their predictive performance was scored using AUROC and AUPRC.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-9254914/v1/83d1ac12cb9f5d18746719fa.png"},{"id":107614288,"identity":"7bcbee35-6246-4a92-a777-3099139c1d07","added_by":"auto","created_at":"2026-04-23 09:07:56","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":812478,"visible":true,"origin":"","legend":"\u003cp\u003eTeam Spider Bobs’ workflow for Task 1 – Classification challenge.\u003cem\u003e\u003cbr\u003e\n \u003c/em\u003e\u0026nbsp;(a) Benchmark model: Key features were selected to train a multi-layer perceptron (MLP).\u003cbr\u003e\n \u0026nbsp;(b) A correlation network analysis approach was used to identify the most important features in the Health \u0026amp; Exposure, Internal Exposome, and External Exposome Survey data:\u003c/p\u003e\n\u003cp\u003ei. Feature selection was based on \u003cem\u003eP\u003c/em\u003e-values and odds ratios from PEGS ExWAS results.\u003c/p\u003e\n\u003cp\u003eii. Correlations among features were obtained from PEGS correlation globes, and a correlation network was created (nodes = features, edges = correlation value).\u003c/p\u003e\n\u003cp\u003eiii. Subnetworks were created from the correlation network for each survey with features selected from PEGS ExWAS results. For correlations, edges \u0026nbsp;\u0026gt; 0.6 were filtered. From each subnetwork, one representative feature was selected from each connected component (including isolated nodes). For model training, three of these features were eventually selected in addition to the features included in the benchmark model. In each case, features with a higher odds ratio were selected. Due to data availability, the model was divided into two parts. Two MLP classifiers were trained on the two data subsets and used for disease prediction (AUROC = 0.7412).\u003c/p\u003e\n\u003cp\u003e(c) Final model: Features with the highest significance were selected based on a literature review, data processing, and hyperparameter optimization. Two gradient boosting classifiers were used for training due to data availability.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-9254914/v1/4b16666d4dc2180c3b8315c9.png"},{"id":107614290,"identity":"ccbea79b-b16b-4762-8e38-95b26af74f31","added_by":"auto","created_at":"2026-04-23 09:07:56","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":500695,"visible":true,"origin":"","legend":"\u003cp\u003eTeam Nonsense-Mediated Decay’s model for Task 1– Classification challenge. The hybrid model integrated predictors from both genomic and survey data.\u003c/p\u003e\n\u003cp\u003e(a) A random forest classifier was used for model training with Health \u0026amp; Exposure Survey data to determine hypercholesterolemia probability. For individuals with whole-genome sequencing (WGS) data, 12 polygenic scores (PGS) associated with hypercholesterolemia were calculated using harmonized weight sets from the PGS Catalog.\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-9254914/v1/a758cf89066eb04dd3e64b84.png"},{"id":108008460,"identity":"bf1b6b68-b3f7-4d8b-aa03-e37fdc7f9e10","added_by":"auto","created_at":"2026-04-28 13:06:47","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3289007,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9254914/v1/3670fe35-ec2f-4c42-a766-bb2d59f3aa8d.pdf"},{"id":107614284,"identity":"dc44992b-2e92-44ac-82bd-0e68c16716ad","added_by":"auto","created_at":"2026-04-23 09:07:56","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":22687,"visible":true,"origin":"","legend":"","description":"","filename":"PEGSDREAMChallengemanuscriptsupplement26092025.docx","url":"https://assets-eu.researchsquare.com/files/rs-9254914/v1/911b45a20e98d77e4d566f6f.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"The PEGS DREAM Challenge: A Crowdsourcing Approach to Understanding Hypercholesterolemia with Multi- dimensional Genomic and Environmental Data","fulltext":[{"header":"Introduction","content":"\u003cp\u003eRecognizing the profound influence of gene-environment (GxE) interactions on human health and disease, researchers are increasingly integrating genetic and environmental data in studies of etiology. Hypercholesterolemia, commonly called high cholesterol, is a highly prevalent condition and carries a significant risk for cardiovascular disease \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. This condition exemplifies the complexity of GxE interactions and their impacts on disease \u003csup\u003e\u003cspan additionalcitationids=\"CR3\" citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. While genetic predisposition contributes to individual susceptibility to high cholesterol, lifestyle choices, dietary habits, and environmental exposures also influence hypercholesterolemia status.\u003c/p\u003e \u003cp\u003eTraditional approaches often analyze individual data types in isolation, limiting their ability to capture the full spectrum of disease-driving factors. The recent focus on high-dimensional datasets containing multi-omics data, or data from multiple biological systems, is enabling researchers to elucidate the many factors influencing complex conditions such as hypercholesterolemia. For example, incorporating information on environmental factors with genomic data can reveal how risk alleles interact with specific environmental exposures to modulate disease risk. Integrating diverse data types supports a holistic understanding of disease mechanisms, paving the way for more accurate risk prediction, targeted interventions, and personalized medicine.\u003c/p\u003e \u003cp\u003eDialogue for Reverse Engineering Assessment and Methods (DREAM) Challenges (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://dreamchallenges.org/\u003c/span\u003e\u003cspan address=\"https://dreamchallenges.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), in partnership with Sage Bionetworks, are aimed at translating the \u0026ldquo;wisdom of the crowd\u0026rdquo; into practice with potentially significant impacts on science and human health. The Challenges engage the scientific community in collaborative problem-solving to address fundamental biomedical issues such as disease classification and prediction using innovative computational models. These global Challenges bring together researchers and data scientists from multiple disciplines to develop and benchmark informatic algorithms. In the DREAM Challenge framework, teams address a specific research question by preparing relevant data and developing models, which are then evaluated and shared. DREAM Challenges have addressed a range of diseases, and results have been published in numerous academic journals \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eA DREAM Challenge using data from the Personalized Environment and Genes Study (PEGS), a diverse North Carolina-based cohort of nearly 20,000 individuals, was launched to promote innovative statistical and data science approaches integrating multi-dimensional environmental, genomic, and geospatial data to dissect the etiology of hypercholesterolemia (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The PEGS cohort is described in detail on the PEGS website (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.niehs.nih.gov/research/atniehs/labs/crb/studies/pegs\u003c/span\u003e\u003cspan address=\"https://www.niehs.nih.gov/research/atniehs/labs/crb/studies/pegs\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eExposomic data gathered through comprehensive surveys capture PEGS participants\u0026rsquo; endogenous and exogenous exposures throughout life, including chemical and environmental exposures encountered at work and home, medications, dietary habits, and lifestyle choices. This comprehensive environmental data enables researchers to consider exposures\u0026rsquo; cumulative effects and interactions with genomic factors. The whole-genome sequencing (WGS) data available for the PEGS cohort comprises single nucleotide variants (SNVs), structural variants, human leukocyte antigen (HLA) genotypes, telomeric content, ancestry estimations, and genome-wide methylation profiles (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e), offering insights into both common and rare genetic variations and their potential functional consequences.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eGeospatial data were created by linking PEGS participants\u0026rsquo; addresses to databases that include proximity to contaminant sources, area-level air pollutant concentrations, and social determinants of health indices such as the CDC/ATSDR Social Vulnerability Index (SVI) (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.atsdr.cdc.gov/placeandhealth/svi/index.html\u003c/span\u003e\u003cspan address=\"https://www.atsdr.cdc.gov/placeandhealth/svi/index.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and the Environmental Justice Index (EJI) (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.atsdr.cdc.gov/placeandhealth/eji/index.html\u003c/span\u003e\u003cspan address=\"https://www.atsdr.cdc.gov/placeandhealth/eji/index.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). These indices provide valuable insights into the spatial distribution of environmental risks and their impacts on health disparities.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe PEGS DREAM Challenge (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.synapse.org/PEGS\u003c/span\u003e\u003cspan address=\"https://www.synapse.org/PEGS\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) asked participants to develop novel predictive models for hypercholesterolemia by harnessing the power of this multi-omics dataset, with the aim of more accurate disease classification and risk prediction and improved ability to develop targeted interventions and inform personalized medicine. The Challenge featured distinct classification and ideation tasks. In the classification task (Task 1), teams developed models in the multi-omics PEGS data to classify hypercholesterolemia. To assess the potential value of incorporating exposomic and geospatial data for improving classification accuracy, a polygenic score (PGS) derived solely from WGS data was used as a benchmark. For the ideation task (Task 2), teams generated data-driven hypotheses and/or models on the relationships between the multidimensional PEGS data and high cholesterol risk. The ideation task incentivized outside-the-box thinking and promoted collaboration, with the overall aim of unraveling how GxE interactions affect human health.\u003c/p\u003e"},{"header":"Results","content":"\u003ch2\u003e\u003cstrong\u003ePEGS DREAM Challenge Data\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe PEGS DREAM Challenge (https://www.synapse.org/PEGS) asked participants to leverage the multi-dimensional PEGS dataset to identify factors affecting hypercholesterolemia and thus improve understanding of its etiology. The challenge was launched on May 1, 2024, with a Leaderboard round from May 10 to August 6, 2024, and a Final round from August 6 to August 19, 2024. Ninety participants from five continents\u0026mdash;Asia, Africa, Europe, North America, and Australia\u0026mdash;completed 345 submissions across all rounds of the Challenge. Winners were announced on August 31, 2024 and presented their findings at the Regulatory \u0026amp; Systems Genomics with DREAM Challenges Conference (RSGDREAM2024) in Madison, Wisconsin, USA (https://www.iscb.org/rsgdream2024/home).\u003c/p\u003e\n\u003cp\u003eParticipants took part in one or both tasks:\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003e\u003cstrong\u003eTask 1: Disease Classification:\u003c/strong\u003e Classify individual hypercholesterolemia disease status in a held-out test subset containing health, exposomic, geospatial, and genomic data.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eTask 2: Ideation Challenge:\u003c/strong\u003e Develop novel hypotheses and models using the multi-dimensional PEGS data to improve understanding of hypercholesterolemia etiology beyond conventional clinical and genetic risk factors.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eParticipants were provided questionnaire-based health and exposure, geospatial, and genomic data from PEGS Data Freeze 3.1. Table 4 provides details.\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003e\u003cstrong\u003eQuestionnaire-based data:\u003c/strong\u003e Demographic, health, environmental exposure, socioeconomic status, and lifestyle data collected from three surveys administered to PEGS participants: the Health \u0026amp; Exposure Survey (N = 9,449), the Internal Exposome Survey (N = 3,071), and the External Exposome Survey (N = 3,618).\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eGenomic data:\u003c/strong\u003e WGS data (N=4,737), including SNVs and indel genotypes, structural variant calls, HLA genotypes, aggregate telomeric content, ancestry estimations, and genome-wide methylation profiling data (N=4,724). Methylation profiling was done using the Infinium MethylationEPIC v1.0 BeadChip Kit.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eGeospatial data:\u003c/strong\u003e Exposure estimates and proximity to hazards calculated using geospatial linkages (e.g., proximity to contaminant sources, air pollutant concentrations), Modern Era Retrospective analysis for Research and Applications (MERRA-2) climate/environmental estimates, and SVI and EJI data.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe Challenge data (N = 9,449) were filtered to remove outliers, related individuals, and race mismatches, resulting in a dataset comprising 9,184 individuals. The filtered data were split into individuals with and without WGS data. Each subset was then randomly split into thirds to create training (n = 3,062), validation (n = 3,062), and test (n = 3,060) subsets. Each subset comprised approximately 50% individuals with sequencing data and 50% without sequencing data. The test subset was not shared with Challenge participants and was held out to score Final round submissions. Figure 4 shows the workflow for Task 1.\u003c/p\u003e\n\u003cp\u003eTo protect the privacy of PEGS participants, the original PEGS data were not provided to Challenge participants. Synthetic data, created using the synthpop R library, was provided for model construction and code development. The synthetic data mirrored the structure and sample sizes of the original dataset but did not preserve inter-table correlations. Submitted models were evaluated with the original PEGS data. Privacy measures included de-identification and anonymization with generated IDs in addition to limiting access to original data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eChallenge Task 1: Disease Classification\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn Task 1, participants developed models in the training and validation datasets to classify hypercholesterolemia status in the held-out test dataset. Participants were encouraged to surpass a benchmark area under the receiver operating characteristic curve (AUROC) of 0.7358, obtained from a hypercholesterolemia PGS. This benchmark PGS was computed for PEGS participants using weights and metadata from the hypercholesterolemia PGS by Weissbrod et al. \u003csup\u003e6\u003c/sup\u003e in the Polygenic Score Catalog \u003csup\u003e7,8\u003c/sup\u003e. This PGS was selected due to its inclusion of multiple ancestries and a large number of genomic variants. Model accuracy was assessed using AUROC, with area under the precision-recall curve (AUPRC) as a tiebreaking metric. Teams Spider Bobs (AUROC = 0.7933) and Nonsense-Mediated Decay (AUROC = 0.7919) submitted the top-performing submissions for Task 1.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e\u003cem\u003eTeam Spider Bobs (Task 1)\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eTeam Spider Bobs integrated data from the Health \u0026amp; Exposure Survey and Internal Exposome Survey using random forest and gradient boosting classifiers from Python\u0026rsquo;s scikit-learn library \u003csup\u003e9\u003c/sup\u003e. For data processing, some variables were manually preprocessed. For ordinal variables, coded responses were reordered to reflect ordinality (e.g., Health \u0026amp; Exposure Survey - current physical health rating compared to five years ago as better, worse, or about the same). For continuous (height, weight, age, BMI) and some ordinal (Health \u0026amp; Exposure Survey questions in the fatigue section) variables, missing responses were imputed with the mean value. For skipped questions (e.g., smoking details for non-smokers), missing responses were replaced with appropriate values (e.g., 0). Gradient boosting had better performance on the team\u0026rsquo;s cleaned data and was used for the final model. The team optimized hyperparameters using cross-validation. Reducing the number of features and sub-samples significantly improved predictions.\u003c/p\u003e\n\u003cp\u003eTeam Spider Bobs used the following workflow (Figure 5):\u003c/p\u003e\n\u003col start=\"1\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eBenchmark Model (Figure 5a):\u003c/strong\u003e An initial model using key features from the Health \u0026amp; Exposure survey (weight, height, sex, age) and trained with a multi-layer perceptron (MLP) had an AUROC of 0.7356 in the validation data.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eRefinement (Figure 5b):\u003c/strong\u003e The model was refined by leveraging PEGS exposome-wide association studies (ExWAS) results and correlation globes (https://pegsexplorer.niehs.nih.gov) to identify relevant features based on \u003cem\u003eP\u003c/em\u003e-values and correlations \u003csup\u003e10,11\u003c/sup\u003e. Features were selected based on ExWAS \u003cem\u003eP\u003c/em\u003e-values and odds ratios, and correlations were checked using correlation globes. For each survey data component outlined in Table 1 (except demographic and administrative data), features were connected if the correlation value was \u0026gt; 0.6, and a representative feature from each component was selected (Figure 5). These were further reduced to three main features plus benchmark features to mitigate overfitting. Data were partitioned based on the availability of Internal Exposome Survey data, and two MLP models were trained and applied. This refined approach improved the AUROC to 0.7412.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eFinal Model (Figure 5c):\u003c/strong\u003e For feature selection to identify the most significant features for the final model, a literature review, data processing, and hyperparameter optimization were utilized. Data were split again, and two gradient boosting classifiers were trained.\u0026nbsp;\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch3\u003e\u003cstrong\u003e\u003cem\u003eTeam Nonsense-Mediated Decay (Task 1)\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eTeam Nonsense-Mediated Decay integrated Health \u0026amp; Exposure Survey and genetic data to enhance classification accuracy. Decision trees and ensemble methods such as random forest and XGBoost were used to capture non-linear relationships between mixed data types (e.g., numerical and categorical variables) and perform feature selection. For genomic data, PGS were utilized to complement the random forest model and capture the combinations of variants influencing disease risk. Data were processed to limit analysis to numeric and categorical variables from the Health \u0026amp; Exposure Survey. Variables with more than 5% missing values were removed; remaining missing data were imputed using the median for numeric data and mode for categorical data.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTeam Nonsense-Mediated Decay used the following workflow for a random forest classifier with Health \u0026amp; Exposure Survey data combined with PGS: (Figure 6):\u0026nbsp;\u003c/p\u003e\n\u003col start=\"1\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eBenchmark Model:\u003c/strong\u003e An initial model was trained with a random forest classifier using only Health \u0026amp; Exposure Survey data to estimate baseline probabilities of hypercholesterolemia status, leveraging the model\u0026rsquo;s inherent feature selection.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eRefinement:\u003c/strong\u003e Twelve PGS associated with hypercholesterolemia were calculated for PEGS participants with genetic data using harmonized weight sets from the PGS Catalog \u003csup\u003e7\u003c/sup\u003e. These PGS and the baseline random forest predictions were combined using logistic regression to refine individual disease probabilities.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eFinal Model:\u003c/strong\u003e The final probability of hypercholesterolemia for each individual \u003cem\u003ek\u003c/em\u003e was computed using the formula:\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003e\u003cimg src=\"https://myfiles.space/user_files/58895_8739fc6c57c1c19a/58895_custom_files/img1776865318.png\" width=\"839\" height=\"138\"\u003e\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003ewhere \u003cem\u003eP̂(Y\u003csub\u003ek\u003c/sub\u003e)\u003c/em\u003e is the predicted probability of hypercholesterolemia for individual \u003cem\u003ek\u003c/em\u003e, \u003cem\u003eX\u003csub\u003ek,b\u003c/sub\u003e\u003c/em\u003e is the baseline probability estimated by the random forest model, \u003cem\u003eX\u003c/em\u003e\u003csub\u003ek,gi\u003c/sub\u003e is the \u003cem\u003ei-\u003c/em\u003eth PGS for individual \u003cem\u003ek\u003c/em\u003e, and \u0026beta; represents the logistic regression coefficients. This two-step integration optimized and regularized the weight of genetic and survey predictors. The hybrid model achieved improved classification performance (AUROC=0.775 on the validation dataset) with respect to its individual components using survey data or genetic data only (with AUROCs of 0.758 and 0.736, respectively), demonstrating the utility of combining multimodal data. By incorporating PGS, the model leveraged insights from previous large-scale studies to interpret genome-wide variation.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eChallenge Task 2: Ideation Challenge\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eIn Task 2, participants leveraged the multi-dimensional PEGS dataset to develop novel models and hypotheses with the aim of improving understanding of hypercholesterolemia etiology beyond conventional genetic and clinical factors. Judges scored the hypotheses and models based on creativity, significance, interpretability, innovation, utility, feasibility, and potential translational impact. As with Task 1, the top-performing submissions for Task 2 were from Teams \u003cstrong\u003eSpider Bobs\u003c/strong\u003e and \u003cstrong\u003eNonsense-Mediated Decay\u003c/strong\u003e.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e\u003cem\u003eTeam Spider Bobs (Task 2)\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eFor Task 2, Team Spider Bobs conducted follow-up analyses from their Task 1 model to identify novel factors influencing hypercholesterolemia risk. Methylation data were analyzed using the \u003cem\u003elimma\u0026nbsp;\u003c/em\u003eR package \u003csup\u003e12\u003c/sup\u003e to detect significant differentially methylated probes (adjusted \u003cem\u003eP\u003c/em\u003e-value \u0026lt; 0.05) for individuals with and without hypercholesterolemia. The results revealed 10 probes (seven hyper, three hypo) corresponding to three hypermethylated genes (\u003cem\u003eELOVL2\u003c/em\u003e, \u003cem\u003eFHL2\u003c/em\u003e, \u003cem\u003eZYG11A\u003c/em\u003e) and two hypomethylated genes (\u003cem\u003ePXN\u003c/em\u003e, \u003cem\u003eCCDC102B\u003c/em\u003e). \u003cem\u003eELOVL2\u003c/em\u003e is involved in fatty acid elongation, potentially influencing lipid metabolism, and \u003cem\u003eFHL2\u003c/em\u003e has been linked with cardiovascular disease.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNext, association analyses were performed for the HLA genotype data using PLINK\u0026rsquo;s \u003csup\u003e13\u003c/sup\u003e logistic regression model with quality control steps that included removing samples/SNPs with excessive missing data, rare variants, and deviations from Hardy-Weinberg equilibrium. There was a significant difference in the distribution of \u003cem\u003eHLA-DPB1\u003c/em\u003e alleles between individuals with and without hypercholesterolemia (\u0026chi;\u0026sup2; = 60.01, p = 0.0072). This association remained significant after multiple testing correction. Marginally significant associations with \u003cem\u003eHLA-H\u003c/em\u003e (\u003cem\u003eP\u003c/em\u003e = 0.0534) and \u003cem\u003eHLA-DMA\u003c/em\u003e (\u003cem\u003eP\u003c/em\u003e = 0.0320) were also observed. There is a known association between familial hypercholesterolemia and certain HLA alleles, and elevated total cholesterol has been correlated with specific HLA variants.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAdditional analysis of features included in the gradient boosting model developed for Task 1 revealed that, in addition to well-known risk factors (e.g., BMI, age, smoking, hypertension, alcohol, poor diet, diabetes), factors related to secondary hypercholesterolemia (i.e., high cholesterol triggered by other diseases), such as uterine tumors, also played an important role. The team identified three novel features significantly associated with hypercholesterolemia status in the PEGS data:\u0026nbsp;\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003eVitamin E taken regularly in the past year (from the Internal Exposome Survey)\u003c/li\u003e\n \u003cli\u003eRegular exposure to dyes (from the Health \u0026amp; Exposure Survey)\u003c/li\u003e\n \u003cli\u003eEver diagnosed with German Measles (from the Internal Exposome Survey)\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eUsing the Informatics for Integrating Biology and the Bedside (i2b2) self-service web-based tool for data exploration (see details in the Methods), Team Spider Bobs further examined these associations. For exposure to dyes, the team hypothesized a correlation with a low standard of living, a known factor for hypercholesterolemia (i.e., workers in chemical/textile manufacturing may earn below-average wages).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSurprisingly, the correlation between diagnosis with German Measles and hypercholesterolemia status had an extremely small \u003cem\u003eP\u003c/em\u003e-value (p \u0026asymp; 6.7e-32). An interaction between the German Measles vaccine and the \u003cem\u003eHLA-DPB1\u003c/em\u003e gene has been reported. A small number of individuals fail to build protective antibodies despite vaccination, and differences in immunity to the Rubella virus have been associated with the \u003cem\u003eHLA-DPB1\u003c/em\u003e gene\u003csup\u003e14,15\u003c/sup\u003e. Based on the observed correlations in PEGS data and improved model scores in Task 1 when including \u0026lsquo;Ever diagnosed with German Measles\u0026rsquo;, the team hypothesized that infection with German Measles is associated with hypercholesterolemia status in the PEGS data.\u003c/p\u003e\n\u003cp\u003eTo test the hypothesis accurately, the team suggested using German Measles antibody results for PEGS participants, as self-reported diagnoses are prone to error (around half of infections go undetected)\u003csup\u003e16\u003c/sup\u003e. This antibody data would enable further testing of the association of hypercholesterolemia and \u003cem\u003eHLA-DPB1\u003c/em\u003e. However, the team acknowledges that this putative association could be due to the potential correlation of German Measles and hypercholesterolemia with socioeconomic factors.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e\u003cem\u003eTeam Nonsense-Mediated Decay (Task 2)\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eFor Task 2, Team Nonsense-Mediated Decay proposed a GxE interaction model that integrates genomic data on SNVs with Health \u0026amp; Exposure Survey data. These components were selected due to their broad availability in the cohort and their observed utility in improving Task 1 classification performance.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eData processing steps include the following. SNV genotypes are encoded based on alternative allele dosage (0, 1, 2). To ensure statistical robustness, only SNVs with a minor allele frequency \u0026gt; 0.05 are considered. Population structure is accounted for with principal component analysis (PCA) of the genotype matrix, with top components included as covariates. Health \u0026amp; Exposure survey responses are filtered, and variables with more than 5% missing responses and more than 95% single response frequency are removed. Variables with strong collinearity or sex-dependent responses are also removed. Numerical and ordered categorical variables are used as-is; binary variables were encoded as 0/1; unordered categorical and free-text variables were excluded.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe marginal effect\u0026nbsp;a\u0026nbsp;of a given SNV (\u003cem\u003eG\u003c/em\u003e) on trait (\u003cem\u003eY\u003c/em\u003e) is first quantified as:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cimg width=\"125\" height=\"19\" src=\"https://myfiles.space/user_files/58895_8739fc6c57c1c19a/58895_custom_files/img1776865356.png\" v:shapes=\"_x0000_i1025\" alt=\"image\"\u003e\u003c/p\u003e\n\u003cp\u003ewhere \u003cem\u003ea\u003c/em\u003e\u003cem\u003e\u003csub\u003e0\u003c/sub\u003e\u003c/em\u003e is the intercept and \u003cimg width=\"13\" height=\"19\" src=\"https://myfiles.space/user_files/58895_8739fc6c57c1c19a/58895_custom_files/img177686535658.png\" v:shapes=\"_x0000_i1025\" alt=\"image\"\u003eis the error term. \u003c/p\u003e\n\u003cp\u003eTo estimate the genetic main effect \u003cem\u003eb\u003c/em\u003e\u003cem\u003e\u003csub\u003e1\u003c/sub\u003e\u003c/em\u003e, in the presence of the environment (\u003cem\u003eE\u003c/em\u003e), the model can be extended to:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cimg width=\"267\" height=\"21\" src=\"https://myfiles.space/user_files/58895_8739fc6c57c1c19a/58895_custom_files/img177686535683.png\" v:shapes=\"_x0000_i1025\" alt=\"image\"\u003e\u003c/p\u003e\n\u003cp\u003ewhere \u003cem\u003eb\u003c/em\u003e\u003cem\u003e\u003csub\u003e2\u003c/sub\u003e\u003c/em\u003e represents the environmental main effect, and \u003cem\u003eb\u003c/em\u003e\u003cem\u003e\u003csub\u003e3\u003c/sub\u003e\u003c/em\u003e the interaction effect. Covariates from PCA and other relevant confounders such as sex are included in the models to account for population structure and confounding.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo detect significant GxE interactions, the linear models described above can be fit to all selected G,E pairs. The relationship between the genetic marginal effects (\u003cem\u003ea\u003c/em\u003e\u003cem\u003e\u003csub\u003e1\u003c/sub\u003e\u003c/em\u003e) can then be modeled in the absence of \u003cem\u003eE\u003c/em\u003e and the genetic main effects (\u003cem\u003eb\u003c/em\u003e\u003cem\u003e\u003csub\u003e1\u003c/sub\u003e\u003c/em\u003e) in the presence of \u003cem\u003eE\u003c/em\u003e with linear regression. As previously demonstrated \u003csup\u003e17\u003c/sup\u003e, in the absence of a true GxE interaction effect, \u003cem\u003eb\u003c/em\u003e\u003cem\u003e\u003csub\u003e1\u003c/sub\u003e\u003c/em\u003e and a\u003csub\u003e1\u003c/sub\u003e are linearly correlated. This developed hypothesis leverages this property to systematically test for the presence of GxE interactions in the data. Potentially interacting G,E pairs can be ranked based on their deviation from this expected correlation and reviewed for biological interpretation.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe PEGS DREAM Challenge demonstrates the power of crowdsourced problem-solving in leveraging multidimensional data to investigate the etiology of a complex condition\u0026mdash;hypercholesterolemia. This initiative brought together diverse expertise from across the world to develop integrative models and testable hypotheses, yielding both methodological innovation and new biological insights. The models presented in the PEGS DREAM Challenge demonstrate that integrative approaches combining self-reported exposome and health survey-based data with genomic information can outperform models based on PGS alone. This finding underscores the value of incorporating multi-dimensional data, such as those in the PEGS cohort, to capture the complex interplay of genetic and environmental factors in diseases like hypercholesterolemia. This Challenge exemplifies how community engagement in a competitive yet collaborative setting can accelerate methodological advances and generate new research directions in precision environmental health.\u003c/p\u003e \u003cp\u003eIn Challenge Task 1, the top-performing teams substantially outperformed the benchmark PGS (AUROC\u0026thinsp;=\u0026thinsp;0.7358), with AUROCs of 0.7933 and 0.7919, respectively, highlighting the usefulness of integrating survey-based health and exposure data with genomic information. Team Spider Bobs relied on correlation networks derived from ExWAS results to identify informative features, while Team Nonsense-Mediated Decay combined ensemble learning with PGS to build a hybrid model. The comparable performance of these two methodologically distinct approaches demonstrates that robust prediction of hypercholesterolemia status can be achieved using flexible combinations of multimodal data. Importantly, these models show that self-reported health and exposure variables\u0026mdash;often underutilized in clinical prediction\u0026mdash;add measurable value to disease classification beyond genetic risk scores.\u003c/p\u003e \u003cp\u003eChallenge Task 2 extended the utility of the PEGS dataset beyond classification, asking participants to generate novel hypotheses regarding hypercholesterolemia risk. These ideation submissions revealed potential biological mechanisms and environmental contributors that warrant further investigation. For example, Team Spider Bobs identified differentially methylated probes in genes such as \u003cem\u003eELOVL2\u003c/em\u003e and \u003cem\u003eFHL2\u003c/em\u003e, implicating lipid metabolism and cardiovascular regulation pathways. Their finding of a significant association between HLA-DPB1 alleles and hypercholesterolemia echoes known links between HLA variation and lipid phenotypes, suggesting immune-mediated contributions to metabolic disease risk. Unexpected associations of hypercholesterolemia with self-reported exposure to dyes and diagnosis with German Measles also emerged. Although these findings may reflect latent confounding (e.g., socioeconomic status), they illustrate how unconventional variables can surface through hypothesis-free exploration and generate new research questions. Team Nonsense-Mediated Decay, meanwhile, proposed a systematic method for detecting GxE interactions using deviations from expected genotype-phenotype correlations in the presence of environmental variables. Their approach, rooted in modeling marginal versus conditional genetic effects, highlights the utility of theoretical expectations to identify non-additive interactions. This strategy is broadly applicable to other traits and datasets, offering a reproducible framework for large-scale GxE analysis.\u003c/p\u003e \u003cp\u003eBeyond the findings themselves, the Challenge showcased key methodological innovations. The use of synthetic data allowed teams to develop and refine models while maintaining participant privacy. The provision of resources such as PEGS correlation globes and ExWAS results enabled participants to perform informed feature selection and contextual interpretation. These tools, combined with Docker-based model evaluation, ensured reproducibility and transparency in model scoring, setting a standard for future data science competitions involving sensitive health data.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eUltimately, the PEGS DREAM Challenge fostered a fertile environment for developing new approaches to integrating genomic, environmental, and social determinants of health data. The top-performing models offered not only improved classification tools but also a roadmap for incorporating diverse data types into disease modeling. These models could be clinically relevant and help build robust and translatable predictive tools for personalized medicine with validation against objective clinical outcomes using data such as electronic health records. The ideation task reinforced the importance of hypothesis generation in complex trait research, particularly when guided by novel, interpretable, and potentially actionable variables. Taken together, the results of the Challenge suggest that integrative, crowdsourced approaches can reveal both methodological and biological insights into diseases shaped by GxE, which will be critical as biobank-scale datasets continue to expand.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"Methods","content":"\u003ch2\u003e\u003cstrong\u003ePEGS DREAM Challenge Data\u003c/strong\u003e\u003c/h2\u003e\n\u003ch3\u003e\u003cstrong\u003e\u003cem\u003ePersonalized Environment and Genes Study (PEGS) cohort\u003c/em\u003e\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eOriginally established in 2002 as the Environmental Polymorphisms Registry (EPR) to recruit participants for ongoing research at NIEHS by convenience sampling at community events, PEGS was renamed in 2022. This racially and ethnically diverse North Carolina-based cohort is a repository of data on medications, health outcomes, environmental exposures, lifestyle factors, genomic data, and geospatial estimates of exposure (Figure 1). PEGS participants complete three surveys\u0026mdash;the Health \u0026amp; Exposure Survey and the Internal and External Exposome Surveys\u0026mdash;and provide biological samples. Participants can consent to be called back for additional tissue collection for add-on studies. Further details of PEGS can be found at \u0026nbsp;https://www.niehs.nih.gov/research/atniehs/labs/crb/studies/pegs/index.cfm. In contrast to typical studies that focus on a single disease or environmental exposure, PEGS gathers data on a wide array of diseases and environmental exposures, including dietary and lifestyle factors, in conjunction with genomic data. The overarching aim of PEGS is to empower researchers to unravel the etiology of disease and uncover how environmental, dietary, lifestyle, and genetic factors collectively impact human health.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe PEGS cohort (N=19,445) comprises approximately two-thirds female (62.3%; 12,120) and one-third male (37.6%; 7,298) participants ranging in age from 18.4 to 98.3 years, with a mean age of 50.2 years (at completion of the Health \u0026amp; Exposure Survey). The self-reported racial makeup of the cohort is two-thirds White (63%; 12,279), slightly over one-quarter Black (27.6%; 5,361), and 4.5% who identify as another race (868). Additionally, 5.0% self-reported their ethnicity as Hispanic (979). This diverse cohort includes participants of varying socioeconomic status and education levels, enabling research on disease risk in multiple populations. Further, this diversity supports broadly applicable results and can help uncover health disparities that occur due to disproportionate environmental exposures for certain populations.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e\u003cem\u003eData provided to PEGS DREAM Challenge teams\u003c/em\u003e\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003ePEGS DREAM Challenge teams were provided access to data from PEGS Data Freeze 3.1 for PEGS participants who completed the Health \u0026amp; Exposure Survey (N = 9,449) (see Table 1). The questionnaire-based data include information on demographics, health, environmental exposures, socioeconomic status, and lifestyle factors (Tables 1 and 2). Genomic data on SNVs and indel genotypes, structural variant calls, HLA genotypes, estimated aggregate telomeric content, and ancestry estimations were derived from WGS. Genome-wide methylation profiling data were obtained using the Infinium MethylationEPIC v1.0 BeadChip Kit (Figure 6). Geospatial data include exposure estimates and proximity to hazards calculated using geospatial linkages with various databases, linkages from the MERRA-2 project providing estimates of climate and environmental metrics from satellite observations, and SVI and EJI data (Table 3).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1.\u003c/strong\u003e PEGS data components. The various data components available in the PEGS cohort. Data reported from PEGS Data Freeze 3.1 created on 6/27/2023.\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"584\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCategory\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eComponent\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDescription\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003eSurvey Data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eDemographic and administrative data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eDemographics, consent, addresses, and administrative data for all participants\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eHealth \u0026amp; Exposure Survey\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eDemographics, health, family history of disease, environmental exposures, socioeconomic status, and lifestyle factors\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eExternal Exposome Survey (Exposome A)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eResidential and occupational environmental exposures\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eInternal Exposome Survey (Exposome B)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eMedication use, physical activity, stress, sleep, diet, genetics, and reproductive history\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003eGeospatial Data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eHazards data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eExposure estimates and proximity measures calculated using geospatial linkages from various databases\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eMERRA-2 Data (Earthdata)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eGeospatial data linkages from the MERRA-2 project containing consistent estimates of climate and environmental metrics from a range of satellite-based environmental observations\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eSocial Vulnerability Index (SVI) data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eGeospatial data linkages for the CDC/ATSDR SVI containing summaries of social determinants of health at the census-tract level\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eEnvironmental Justice Index (EJI) data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eGeospatial data linkage for the CDC/ATSDR EJI containing summaries of environmental, social, and health factors at the census-tract level\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003eGenomic data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eSingle nucleotide variants (SNVs)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eSNV and small indel genotypes derived from the whole genome sequencing (WGS) data in PLINK\u0026rsquo;s .bed/.bim/.fam format\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eStructural variants\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eStructural variant calls generated from the WGS data in .vcf format consisting of large deletions, duplications, and inversions\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eHuman leukocyte antigens (HLA) Genotypes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eHLA genotypes identified from the WGS data for 20 HLA genes with up to six digits of specificity\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eTelomeric content\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eAggregate telomeric content estimated from WGS reads reported as telomeric reads per GC content-matched million reads\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eLocal and global ancestry estimations\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eInferred local ancestry per chromosome after haplotype phasing and global estimates of percent ancestry for each participant\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 15.7534%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1438%;\"\u003e\n \u003cp\u003eMethylation data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60.1027%;\"\u003e\n \u003cp\u003eGenome-wide methylation profiling data using the Infinium MethylationEPIC v1.0 BeadChip Kit targeting 866,297 CpG sites\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2.\u003c/strong\u003e Categories of questions in the PEGS Health \u0026amp; Exposure Survey, External Exposome Survey, and Internal Exposome Survey. The table shows the high-level survey question categories of the surveys administered to PEGS participants. Data reported from PEGS Data Freeze 3.1 created on 6/27/2023.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"624\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\" valign=\"top\" style=\"width: 624px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003cstrong\u003eHealth \u0026amp; Exposure Survey\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 135px;\"\u003e\n \u003cp\u003eAbout Your Family\u0026rsquo;s Health\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 373px;\"\u003e\n \u003cp\u003eDiabetes and Endocrine\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003eNeurologic\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 135px;\"\u003e\n \u003cp\u003eAbout Your General Health\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 373px;\"\u003e\n \u003cp\u003eDigestive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003eOccupation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 135px;\"\u003e\n \u003cp\u003eAbout Your Home Life\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 373px;\"\u003e\n \u003cp\u003eExposures\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003eRenal\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 135px;\"\u003e\n \u003cp\u003eAbout Your Mood\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 373px;\"\u003e\n \u003cp\u003eFatigue\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003eReproductive (Females Only)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 135px;\"\u003e\n \u003cp\u003eBones, Joints, and Muscles\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 373px;\"\u003e\n \u003cp\u003eHematological\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003eReproductive (Males Only)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 135px;\"\u003e\n \u003cp\u003eCancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 373px;\"\u003e\n \u003cp\u003eImmune\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003eRespiratory\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 135px;\"\u003e\n \u003cp\u003eCardiovascular\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 373px;\"\u003e\n \u003cp\u003eLifestyle\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 116px;\"\u003e\n \u003cp\u003eSkin, Eyes, and Hair\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" valign=\"top\" style=\"width: 623px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 1px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\" valign=\"top\" style=\"width: 624px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eExternal Exposome (Exposome A)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"3\" rowspan=\"3\" valign=\"top\" style=\"width: 370px;\"\u003e\n \u003cp\u003eCharacteristics of Current and Past Residences:\u003cbr\u003e\u0026nbsp;\u0026bull; Agricultural Property Use\u003cbr\u003e\u0026nbsp;\u0026bull; Garage and Basement\u003cbr\u003e\u0026nbsp;\u0026bull; Heating and Cooling\u003cbr\u003e\u0026nbsp;\u0026bull; Pesticides and Insecticides\u003cbr\u003e\u0026nbsp;\u0026bull; Pets\u003cbr\u003e\u0026nbsp;\u0026bull; Surrounding Area\u003cbr\u003e\u0026nbsp;\u0026bull; Walls and Flooring\u003cbr\u003e\u0026nbsp;\u0026bull; Water and Dampness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 253px;\"\u003e\n \u003cp\u003eChemical and Metal Exposures at Work\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 1px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 253px;\"\u003e\n \u003cp\u003eUltraviolet Light Exposure\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 1px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 253px;\"\u003e\n \u003cp\u003eWorkplace Characteristics\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 1px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 370px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 254px;\"\u003e\n \u003cp\u003eHobby Exposures\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" valign=\"top\" style=\"width: 623px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 1px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\" valign=\"top\" style=\"width: 624px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eInternal Exposome (Exposome B)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 207px;\"\u003e\n \u003cp\u003eChemotherapy/Radiation Therapy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 417px;\"\u003e\n \u003cp\u003ePhysical Activity\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 207px;\"\u003e\n \u003cp\u003eDietary Behavior\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 417px;\"\u003e\n \u003cp\u003eReproductive History (Females Only)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 207px;\"\u003e\n \u003cp\u003eDietary Intake\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 417px;\"\u003e\n \u003cp\u003eSleep\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 207px;\"\u003e\n \u003cp\u003eGenetic History\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 417px;\"\u003e\n \u003cp\u003eStress\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 207px;\"\u003e\n \u003cp\u003eInfectious Disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 417px;\"\u003e\n \u003cp\u003eVitamins, Minerals, and Other Supplement Use\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 207px;\"\u003e\n \u003cp\u003eMedications\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 417px;\"\u003e\n \u003cp\u003eTwin/Triplet Siblings and Birth Order\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\" valign=\"top\" style=\"width: 624px;\"\u003e\n \u003cp\u003eOther\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3.\u003c/strong\u003e Summary of PEGS geospatial data. An overview of the geocoding and GIS data linkages available in the PEGS cohort. Data reported from PEGS Data Freeze 3.1 created on 6/27/2023.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"593\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003cstrong\u003eSource\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDescription\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eExamples\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eGeocodes (GIS)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eGeocoded data from multiple participant-provided addresses at the time of initial enrollment, completion of the Health \u0026amp; Exposure Survey, completion of the External Exposome Survey, the longest-lived childhood address, and the longest-lived adulthood address from the External Exposome Survey\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eGeographic coordinates (latitude and longitude) from multiple participant-provided addresses\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eHazards\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eExposure estimates computed from Department of Transportation (DOT) data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eInformation on train tracks, rail depots, and roadways, such as total major roadway length and distance to the nearest rail depot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eHazards\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eExposure estimates computed from Federal Aviation Administration (FAA) data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eInformation on aircraft departure and arrival sites (e.g., distance to the nearest airport)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eHazards\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eExposure estimates computed from Federal Communications Commission (FCC) data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eInformation on cellular network towers (e.g., nearest cell tower)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eHazards\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eExposure estimates computed from the North Carolina Department of Environmental Quality (NCDEQ)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eDistance to multi-pollutant point sources such as swine caged feeding operations (CAFOs), hazardous waste sites, hazardous spill sites, EPA superfund sites, and wastewater treatment plant release sites\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eHazards\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eExposure estimates computed from Nuclear Regulatory Commission (NRC) data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eDistance to nuclear power stations\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eHazards\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eExposure estimates computed from Atmospheric Composition and Analysis Group (ACAG) data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eParticulate matter concentrations such as PM2.5 total, PM2.5 sulfate, and PM2.5 black carbon and other\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eHazards\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eExposure estimates computed from Center for Air, Climate, and Energy Solutions (CACES) data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eConcentrations for multiple pollutants such as carbon monoxide, nitrogen dioxide, and ozone concentration\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eHazards\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eExposure estimates computed from Toxics Release Inventory (TRI) data\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eEmissions for chemicals of interest such as benzene, ethylbenzene, xylene, and toluene\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eMERRA-2 data (Earthdata)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eGeospatial data linkages from the Modern Era Retrospective Analysis for Research and Applications (MERRA-2) project to assimilate a range of satellite-based environmental observations into a consistent estimate of climate and environmental metrics\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eParticulate, gas, meteorological, and health-relevant exposure indicators such as dust sedimentation, organic carbon emission bin, SO\u003csub\u003e2\u003c/sub\u003e biomass burning emissions, and sea-level pressure\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eSocial Vulnerability Index (SVI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eGeospatial data linkages for CDC/ATSDR SVI designed to consistently quantify multiple social determinants of health across the United States over time\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eSummaries of social determinants of health at the census-tract level, including an overall index, four component indices (socioeconomic status, household characteristics, racial and ethnic minority status, and housing type/transportation), and source variables used to compute each index component (e.g., poverty, education, overcrowding, access to a vehicle)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 110px;\"\u003e\n \u003cp\u003eEnvironmental Justice Index (EJI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 219px;\"\u003e\n \u003cp\u003eGeospatial data linkages for CDC/ATSDR EJI containing summaries and ranks of the cumulative impacts of environmental injustice on health at the census-tract level\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eRanks for each census tract based on 36 environmental, social, and health factors grouped into 10 domains and three overarching modules: environmental burden, social vulnerability, and health vulnerability\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThe dataset (N = 9,449) was filtered to remove outliers, related individuals determined from WGS data, and individuals with mismatched self-reported and WGS-inferred race \u0026nbsp;(N = 9,184). The filtered data were split into individuals with and without WGS data. Each subset was then randomly split into thirds to create training (n = 3,062), validation (n = 3,062), and test (n = 3,060) subsets. Accordingly, the training, validation, and test subsets each consisted of approximately 1,500 individuals with and 1,500 individuals without sequencing data. The test subset was not shared with Challenge participants and was held out to score Final round submissions. Table 4 provides the demographics of participants included in the DREAM Challenge data.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4.\u0026nbsp;\u003c/strong\u003ePEGS DREAM Challenge participant demographics and data availability. Demographics of participants and survey and WGS data availability in the PEGS DREAM Challenge data in the training, validation, and test subsets. Age was computed at the time of completion of the Health \u0026amp; Exposure Survey. The PEGS DREAM Challenge data was created from PEGS Data Freeze 3.1 created on 6/27/2023.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDemographic variable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTraining data\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003en (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eValidation data\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003en (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTest data\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003en (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTotal N\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e3062 (33.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e3062 (33.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e3060 (33.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSex\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e994 (32.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e993 (32.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e1039 (34)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e2068 (67.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e2069 (67.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e2021 (66)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRace\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eOther\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e133 (4.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e131 (4.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e140 (4.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eBlack or African American\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e696 (22.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e678 (22.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e620 (20.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eWhite\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e2157 (70.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e2179 (71.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e2214 (72.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e76 (2.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e74 (2.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e86 (2.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eEthnicity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eNon-Hispanic/Non-Latino\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e2886 (94.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e2887 (94.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e2905 (94.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eHispanic/Latino\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e119 (3.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e130 (4.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e112 (3.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e57 (1.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e45 (1.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e43 (1.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eEducation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e12th grade or less\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e528 (17.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e493 (16.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e501 (16.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eCollege, technical or vocational\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e887 (29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e915 (29.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e912 (29.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eBachelor\u0026apos;s degree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e828 (27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e811 (26.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e838 (27.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eGraduate or professional degree\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e800 (26.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e819 (26.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e783 (25.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e19 (0.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e24 (0.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e26 (0.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eIncome\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eLess than $20,000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e432 (14.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e400 (13.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e417 (13.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e$20,000 to 49,999\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e896 (29.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e937 (30.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e907 (29.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e$50,000 to 79,999\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e740 (24.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e728 (23.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e718 (23.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e$80,000 or more\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e892 (29.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e889 (29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e906 (29.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e102 (3.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e108 (3.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e112 (3.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge in years (SD)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e50 (16)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e50.4 (15.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e50.2 (16)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHealth \u0026amp; Exposure Survey completed\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eN=\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e3062 (100)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e3062 (100)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e3060 (100)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eExternal Exposome completed\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eNo\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e2001 (65.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1992 (65.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e2018 (65.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e1061 (34.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1070 (34.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e1042 (34.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eInternal Exposome completed\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e2156 (70.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e2151 (70.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e2155 (70.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e906 (29.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e911 (29.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e905 (29.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eWGS available\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e1547 (50.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1547 (50.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e1546 (50.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e1515 (49.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1515 (49.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e1514 (49.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHypercholesterolemia status\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e1018 (33.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1011 (33.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e983 (32.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 245px;\"\u003e\n \u003cp\u003eNo\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 131px;\"\u003e\n \u003cp\u003e2012 (65.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e2001 (65.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e2029 (66.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eIn the test dataset, 983 (32.8%) PEGS participants self-reported a diagnosis of hypercholesterolemia in the Health \u0026amp; Exposure Survey. Table 4 shows the number of PEGS participants available for the specific PEGS data component. In the PEGS DREAM Challenge data, WGS data were available for 4,544 participants, and Internal and External Exposome Survey data were available for 2,722 and 3,173 participants, respectively.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e\u003cem\u003eSynthetic data\u003c/em\u003e\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eThe synthpop v1.8 library in R was used to create synthetic data from the original tabular PEGS data. For each data file stratified into training/validation/testing sets, a synthetic version equal in size was created using a random forest model, sequentially for each feature. By generating synthetic data for each table separately, inter-table correlation was not preserved. However, the data were appropriate for developing methods that were validated during this Challenge phase. For large survey and genomic data tables, independent resampling and error injection were performed to match the data types in the original dataset but remove correlation between features.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe synthetic dataset contained training (_train.*) and validation (*_val.*) data that reflect the structure and sample sizes of the original PEGS dataset. In the synthetic dataset, \u0026ldquo;NA\u0026rdquo; was used for unmeasured missing phenotype variables to preserve the original data structure. The synthetic and original datasets contain the same number of individuals in each file.\u003c/p\u003e\n\u003cp\u003eMultiple measures were taken to protect the privacy of PEGS participants. First, all data were de-identified by removing all personally identifiable information (PII) and protected health information (PHI). Second, the data were anonymized by replacing participant IDs with generated IDs for each data component. Third, Challenge participants were not provided access to the original PEGS data. Only the synthetic data were available for download for model construction and code development. The submitted models were evaluated using the original PEGS data that were unavailable for download by Challenge participants.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eChallenge Details\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe PEGS DREAM Challenge invited teams to develop models to identify the myriad factors affecting hypercholesterolemia and improve understanding of the etiology of this complex disease. Determining modifiable factors can help reduce the likelihood of developing hypercholesterolemia and its associated health complications. The Challenge was launched on May 1, 2024, and a Leaderboard round ran from May 10 to August 6, 2024. The Final round ran from August 6 to August 19, 2024. Winners were announced on August 31, 2024 and presented their models and findings at the Regulatory \u0026amp; Systems Genomics with DREAM Challenges Conference (RSGDREAM2024) in Madison, Wisconsin, USA (https://www.iscb.org/rsgdream2024/home).\u003c/p\u003e\n\u003cp\u003eChallenge teams could take part in one or both distinct tasks. The goal of Task 1was to classify individual hypercholesterolemia disease status in the held-out test dataset using combinations of the health, exposomic, geospatial, and genomic data available for the PEGS cohort. The goal of Task 2, an ideation challenge, was to develop novel hypotheses and models using the multi-dimensional PEGS data to improve understanding of the etiology of hypercholesterolemia beyond conventional clinical and genetic risk factors.\u003c/p\u003e\n\u003cp\u003eTo ensure rigorous evaluation while safeguarding participant privacy, the Challenge was implemented using a model-to-data framework \u003csup\u003e18\u003c/sup\u003e.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eIn this paradigm, the PEGS cohort data were stored securely on the Challenge platform, and participants submitted containerized models that were executed against hidden validation datasets. This approach, successfully pioneered in previous DREAM Challenges, enabled participants to develop innovative methods without direct access to confidential health and genomic data while organizers ensured unbiased evaluation. This framework was used in previous Challenges that include the Digital Mammography DREAM Challenge (2017), Patient Mortality Prediction EHR DREAM Challenge (2019), COVID-19 EHR DREAM Challenge (2020), and CD2H NLP Sandbox (2021).\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e\u003cem\u003eChallenge Task 1 \u0026ndash; Disease Classification\u003c/em\u003e\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eFigure 4 shows the workflow for Task 1 of the PEGS DREAM Challenge. Participants were provided with synthetic data with the same format and structure as the original PEGS data. For details, see the synthetic data section. The training and validation subsets were available for model training and model optimization, respectively, and participants could combine or split the subsets in various proportions. The test dataset was not shared with challenge contributors and was used for scoring and evaluation.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eParticipants were asked to integrate the multi-dimensional components of PEGS, comprising health, exposure, genomic, and geospatial data, to create a model that surpassed the classification accuracy of an AUROC = 0.7358. The AUROC value was obtained from a model using a hypercholesterolemia PGS computed for PEGS participants using the weights and metadata from the hypercholesterolemia PGS by Weissbrod et al. \u003csup\u003e6\u003c/sup\u003e from the Polygenic Score Catalog \u003csup\u003e7,8\u003c/sup\u003e as the main predictor. This hypercholesterolemia PGS was chosen because it included the highest number of genomic variants in the score compared to other hypercholesterolemia scores in the catalog and included individuals from multiple ancestries in the evaluation set, mirroring the diversity of the PEGS cohort. A detailed description of PGS computation for PEGS participants is provided in Schaid et al. \u003csup\u003e19\u003c/sup\u003e \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTeams uploaded Docker containers and writeups in .docx or .pdf format to their Synapse Project workspace. The submitted Docker container produced a single output file classifying hypercholesterolemia status for the held-out test dataset in a two-column .csv file containing participant IDs and disease probability. The submitted Docker containers were run on the original PEGS data, with a limit of five model submissions per day in the Leaderboard round and one model submission per day in the Final round. Scoring was based on the AUROC of each submitted model in the validation subset for the Leaderboard round and in the test dataset for the Final round. AUPRC was employed as a tie-breaking metric.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e\u003cem\u003eChallenge Task 2 \u0026ndash; Ideation Challenge\u003c/em\u003e\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eIn Task 2, Challenge participants leveraged the comprehensive PEGS dataset to develop bold and novel hypotheses and models, integrating its health, exposure, geospatial, and genomic components to provide multi-dimensional insights. This challenge was aimed at harnessing collective creativity to generate a pipeline of fresh ideas to address complex challenges. Participants proposed novel research questions and methodologies, with the goals of fostering ideation related to emerging omics technologies and their applications, incentivizing outside-the-box thinking beyond established analytical approaches, and promoting collaboration and interdisciplinary approaches for holistic exploration.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTeams submitted a writeup in .docx or .pdf format documenting their hypotheses and models, which were scored by a panel of judges based on creativity, significance, interpretability, innovation, utility, feasibility, and potential translational impact. Submissions with working or detailed models or specific mathematical model descriptions were given higher scores. Participants were encouraged to submit models involving GxE interactions, genome-wide environment interaction studies (GWEIS), and similar methods. Innovation was emphasized to encourage participants to explore novel approaches.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e\u003cem\u003eChallenge Resources\u0026nbsp;\u003c/em\u003e\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eThe PEGS DREAM Challenge website (https://www.synapse.org/PEGS) describes in detail the PEGS cohort, data available to participants, and two Challenge tasks and provides submission guidelines and evaluation criteria. The website also outlines the Challenge timeline, participation criteria, and conditions for data use. Additionally, the AUROCs of the models submitted for Task 1 computed in the validation data were published live on the Challenge website during the Leaderboard round to provide participants with feedback and enable them to improve their models for the Final round.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn addition to the Challenge website, participants had access to several other PEGS resources for data exploration, hypothesis generation, and investigation of results of prior analyses in the PEGS data. The PEGS Explorer (https://pegsexplorer.niehs.nih.gov/) web application shares published results of ExWAS conducted in PEGS data and visualizations of the complex correlations among the exposures through correlation globes \u003csup\u003e10,11\u003c/sup\u003e. Participants could explore and use these results for the Challenge. Participants could also request access to the i2b2 self-service web-based tool for data exploration, which enabled participants to explore de-identified and aggregated PEGS data by building queries. The results assisted with data exploration for model construction and hypothesis generation. Participants were also provided with access to code libraries for the PEGS data as a reference for development efforts. This included a library containing common utilitarian functions for ingesting and analyzing the PEGS data (https://github.com/fsakhtari/PEGS_common/blob/master/pegs_common_utils.R) and example scripts (https://github.com/nathanielmacnell/PEGStools) for working with the PEGS geospatial data.\u003c/p\u003e\n\u003cp\u003eAll methods were carried out in accordance with relevant guidelines and regulations. The PEGS study protocol and/or the use of PEGS data for this study were approved by the NIEHS IRB, protocol 04-E-0053. Informed consent was obtained from all participants and/or their legal guardians.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was funded in part by the intramural research program of the National Institute of Environmental Health Sciences.\u0026nbsp;The contributions of the NIH authors were made as part of their official duties as NIH federal employees, are in compliance with agency policy requirements, and are considered Works of the United States Government. However, the findings and conclusions presented in this paper are those of the author(s) and do not necessarily reflect the views of the NIH or the U.S. Department of Health and Human Services.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eF.S.A., The PEGS DREAM Challenge Community, G.S.A, and A.A.M-R. conceptualized the study. F.S.A., J.F., J.J., V.V., E.C., A.S-Y., M-T.H., A.L., F.M., G.A., S.M., V.C., J.A.,D.C.F., C.P.S., and J.E.H curated the data used in the study and performed the formal analysis. F.S.A. and A.A.M-R. provided project administration and wrote the original manuscript draft. G.S.G. supervised the study. F.S.A. and G.S.G. validated the study results. J.F., J.J., V.V., E.C., A.S-Y., M-T.H., A.L., F.M., G.A., S.M., V.C., J.A., D.C.F., C.P.S., and J.E.H reviewed and edited the manuscript. A.A.M-R. acquired funding for the study and was involved in investigation and methodology. All authors approved the submitted version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe would like to thank the PEGS participants for their contributions to this work. We would also like to express our sincere appreciation to Sharon Soucek in the Office of Technology Transfer at NIEHS for support and expertise regarding the data use agreements that enable data sharing and collaborative research projects with PEGS. We would like to thank Sage Bionetworks for hosting the PEGS DREAM Challenge on their platform and providing the necessary infrastructure support. We also thank Hannah Collins Cakar for manuscript support.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe PEGS dataset analyzed in this study is not publicly available because it contains sensitive human participant data and is subject to ethical and privacy restrictions. De-identified data supporting the findings of this study are available from the corresponding author upon reasonable request, subject to review and approval by the NIEHS IRB, compliance with applicable ethical and legal requirements, and execution of any required data use agreement. Requests for access should be directed to [email protected] or through a web form inquiry: https://www.niehs.nih.gov/research/atniehs/labs/crb/studies/pegs/collaboration/proposal.\u003c/p\u003e\n\u003cp\u003ePEGS DREAM Challenge participants were required to provide code as Docker containers to run their models for evaluation. Code libraries for working with PEGS data were provided as resources (e.g., https://github.com/fsakhtari/PEGS_common/blob/master/pegs_common_utils.R, https://github.com/nathanielmacnell/PEGStools).\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWong, N. D., Lopez, V., Tang, S. \u0026amp; Williams, G. R. Prevalence, treatment, and control of combined hypertension and hypercholesterolemia in the United States. \u003cem\u003eAm J Cardiol\u003c/em\u003e \u003cstrong\u003e98\u003c/strong\u003e, 204\u0026ndash;208 (2006). https://doi.org/https://doi.org/10.1016/j.amjcard.2006.01.079\u003c/li\u003e\n\u003cli\u003eOrdovas, J. M.\u003cem\u003e et al.\u003c/em\u003e Gene-diet interaction in determining plasma lipid response to dietary intervention. \u003cem\u003eAtherosclerosis\u003c/em\u003e \u003cstrong\u003e118 Suppl\u003c/strong\u003e, S11\u0026ndash;27 (1995). \u003c/li\u003e\n\u003cli\u003eOrdovas, J. M. \u0026amp; Shen, J. Gene\u0026ndash;environment interactions and susceptibility to metabolic syndrome and other chronic diseases. \u003cem\u003eJ Periodontol\u003c/em\u003e \u003cstrong\u003e79\u003c/strong\u003e, 1508\u0026ndash;1513 (2008). \u003c/li\u003e\n\u003cli\u003eYe, S. Q. \u0026amp; Kwiterovich, P. O., Jr. Influence of genetic polymorphisms on responsiveness to dietary fat and cholesterol. \u003cem\u003eAm J Clin Nutr\u003c/em\u003e \u003cstrong\u003e72\u003c/strong\u003e, 1275s\u0026ndash;1284s (2000). https://doi.org/10.1093/ajcn/72.5.1275s\u003c/li\u003e\n\u003cli\u003eMeyer, P. \u0026amp; Saez-Rodriguez, J. Advances in systems biology modeling: 10 years of crowdsourcing DREAM challenges. \u003cem\u003eCell Syst\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 636\u0026ndash;653 (2021). https://doi.org/10.1016/j.cels.2021.05.015\u003c/li\u003e\n\u003cli\u003eWeissbrod, O.\u003cem\u003e et al.\u003c/em\u003e Leveraging fine-mapping and multipopulation training data to improve cross-population polygenic risk scores. \u003cem\u003eNat Genet\u003c/em\u003e \u003cstrong\u003e54\u003c/strong\u003e, 450\u0026ndash;458 (2022). \u003c/li\u003e\n\u003cli\u003eLambert, S. A.\u003cem\u003e et al.\u003c/em\u003e Enhancing the Polygenic Score Catalog with tools for score calculation and ancestry normalization. \u003cem\u003eNat Genet\u003c/em\u003e \u003cstrong\u003e56\u003c/strong\u003e, 1989\u0026ndash;1994 (2024). https://doi.org/10.1038/s41588-024-01937-x\u003c/li\u003e\n\u003cli\u003eLambert, S. A.\u003cem\u003e et al.\u003c/em\u003e The Polygenic Score Catalog as an open database for reproducibility and systematic evaluation. \u003cem\u003eNat Genet\u003c/em\u003e \u003cstrong\u003e53\u003c/strong\u003e, 420\u0026ndash;425 (2021). https://doi.org/10.1038/s41588-021-00783-5\u003c/li\u003e\n\u003cli\u003ePedregosa, F.\u003cem\u003e et al.\u003c/em\u003e Scikit-learn: Machine learning in Python. \u003cem\u003eJ Mach Learn Res\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 2825\u0026ndash;2830 (2011). \u003c/li\u003e\n\u003cli\u003eLloyd, D.\u003cem\u003e et al.\u003c/em\u003e Questionnaire-based exposome-wide association studies for common diseases in the Personalized Environment and Genes Study. \u003cem\u003eExposome\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, osae002 (2024). \u003c/li\u003e\n\u003cli\u003eLloyd, D.\u003cem\u003e et al.\u003c/em\u003e Interactive data sharing for multiple questionnaire-based exposome-wide association studies and exposome correlations in the Personalized Environment and Genes Study. \u003cem\u003eExposome\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, osae003 (2024). \u003c/li\u003e\n\u003cli\u003eRitchie, M. E.\u003cem\u003e et al.\u003c/em\u003e limma powers differential expression analyses for RNA-sequencing and microarray studies. \u003cem\u003eNucleic Acids Res\u003c/em\u003e \u003cstrong\u003e43\u003c/strong\u003e, e47\u0026ndash;e47 (2015). \u003c/li\u003e\n\u003cli\u003eChang, C. C.\u003cem\u003e et al.\u003c/em\u003e Second-generation PLINK: rising to the challenge of larger and richer datasets. \u003cem\u003eGigascience\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, s13742\u0026ndash;13015\u0026ndash;10047\u0026ndash;13748 (2015). \u003c/li\u003e\n\u003cli\u003eKennedy, R. B., Ovsyannikova, I. G., Vierkant, R. A., Jacobson, R. M. \u0026amp; Poland, G. A. Effect of human leukocyte antigen homozygosity on rubella vaccine\u0026ndash;induced humoral and cell-mediated immune responses. \u003cem\u003eHum Immun\u003c/em\u003e \u003cstrong\u003e71\u003c/strong\u003e, 128\u0026ndash;135 (2010). \u003c/li\u003e\n\u003cli\u003eLambert, N. D.\u003cem\u003e et al.\u003c/em\u003e Polymorphisms in HLA-DPB1 are associated with differences in rubella virus\u0026ndash;specific humoral immunity after vaccination. \u003cem\u003eJ Infect Dis\u003c/em\u003e \u003cstrong\u003e211\u003c/strong\u003e, 898\u0026ndash;905 (2015). \u003c/li\u003e\n\u003cli\u003eH\u0026uuml;bschen, J.\u003cem\u003e et al.\u003c/em\u003e Challenges of measles and rubella laboratory diagnostic in the era of elimination. \u003cem\u003eClin Microbiol Infect\u003c/em\u003e \u003cstrong\u003e23\u003c/strong\u003e, 511\u0026ndash;515 (2017). \u003c/li\u003e\n\u003cli\u003eZhu, X.\u003cem\u003e et al.\u003c/em\u003e An approach to identify gene-environment interactions and reveal new biological insight in complex traits. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 3385 (2024). https://doi.org/10.1038/s41467-024-47806-3\u003c/li\u003e\n\u003cli\u003eGuinney, J. \u0026amp; Saez-Rodriguez, J. Alternative models for sharing confidential biomedical data. \u003cem\u003eNat Biotechnol\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, 391\u0026ndash;392 (2018). https://doi.org/10.1038/nbt.4128\u003c/li\u003e\n\u003cli\u003eSchaid, D. J.\u003cem\u003e et al.\u003c/em\u003e Polygenic scores and social determinants of health: Their correlations and potential biases. \u003cem\u003eHum Genet Genom Adv\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e (2025). https://doi.org/10.1016/j.xhgg.2024.100389\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"DREAM Challenge, Personalized Environment and Genes Study, gene-environment interactions, hypercholesterolemia, high cholesterol, crowdsourced data analysis","lastPublishedDoi":"10.21203/rs.3.rs-9254914/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9254914/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eCrowdsourced challenges are powerful catalysts for advancing biomedical research and fostering community-driven innovation. The DREAM Challenges initiative, in collaboration with the Personalized Environment and Genes Study (PEGS), held a competition to spur the development of predictive models for hypercholesterolemia risk. Participants were tasked with integrating diverse data, including environmental exposures, whole-genome sequencing, and geospatial information, from a large and diverse cohort to classify hypercholesterolemia and generate novel insights. The top-performing models, which primarily leveraged gradient boosting and random forest classifiers, demonstrated strong predictive performance, outperforming traditional polygenic scores (PGS). In Challenge Task 1, models from the two top-performing teams substantially outperformed the Challenge benchmark dataset using only PGS (AUROC\u0026thinsp;=\u0026thinsp;0.7358), with AUROCs of 0.7933 and 0.7919, respectively. Beyond prediction, these models highlight the significant value of self-reported health and environmental exposure data, revealing them as informative predictors of hypercholesterolemia risk that complement established clinical and genetic factors. This paper details the design of the challenge, presents the top-performing models, and highlights the potential of integrative, multimodal approaches for understanding and predicting complex human diseases.\u003c/p\u003e","manuscriptTitle":"The PEGS DREAM Challenge: A Crowdsourcing Approach to Understanding Hypercholesterolemia with Multi- dimensional Genomic and Environmental Data","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-23 09:07:45","doi":"10.21203/rs.3.rs-9254914/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-18T18:50:47+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"210145477082901888279624855981722784327","date":"2026-04-16T05:34:59+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-13T18:55:43+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-13T18:52:57+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-04-13T18:47:08+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-06T20:30:45+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-04-06T20:25:14+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"57ee9268-6ec5-4d3e-8be8-e789481adaed","owner":[],"postedDate":"April 23rd, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-18T18:50:47+00:00","index":35,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":66416300,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":66416301,"name":"Biological sciences/Genetics"}],"tags":[],"updatedAt":"2026-04-23T09:07:45+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-23 09:07:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9254914","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9254914","identity":"rs-9254914","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}