Prospective Germline Exome and Machine Learning-Based Risk Score Identify Predictive and PrognosticBiomarkers of Immunotherapy Outcomes in Advanced Non-Small Cell Lung Cancer

Prospective Germline Exome and Machine Learning-Based Risk Score Identify Predictive and PrognosticBiomarkers of Immunotherapy Outcomes in Advanced Non-Small Cell Lung Cancer | 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 Research Article Prospective Germline Exome and Machine Learning-Based Risk Score Identify Predictive and PrognosticBiomarkers of Immunotherapy Outcomes in Advanced Non-Small Cell Lung Cancer Andrea González-Hernández, Alberto Ríos, Juan Luis Onieva, Alexandra Cantero, and 11 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8455214/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 4 You are reading this latest preprint version Abstract Background: Immune checkpoint inhibitors (ICIs) have transformed the treatment landscape of advanced non-small cell lung cancer (NSCLC). However, a substantial proportion of patients do not experience durable clinical benefits, and established tumour biomarkers, such as PD-L1 expression and tumour mutational burden, often show limited predictive value. The potential of inherited germline variants to predict immunotherapy outcomes in NSCLC remains a critical and underexplored area. Methods: We prospectively enrolled 117 patients with advanced NSCLC treated with ICI-based regimens at two centres in Spain. Germline whole-exome sequencing (WES) was performed on pretreatment blood samples. Exonic and intronic variants were annotated and integrated with comprehensive clinical data. We applied XGBoost and LASSO machine learning models to identify predictive germline variants and clinical features, and subsequently trained them to predict treatment response and progression-free survival (PFS). This approach produced a novel clinical–germline risk score, generating both a global model and a specific model for the lung adenocarcinoma (LUAD) histological subtype. Results: XGBoost significantly outperformed penalised regression (LASSO), achieving a robust cross-validated area under the curve (AUC) of 0.845 for predicting treatment response in the validation cohort. Our models identified several novel germline loci that were significantly associated with immunotherapy outcomes. Variants in SLC6A16 , SIGLEC11 , PDE4E , and OR10H5 were associated with reduced PFS, whereas variants in CCZ1B and PHLDB1 were associated with extended PFS. Lymph node metastasis was confirmed as the sole independent clinical predictor of poor response (OR 2.07, P=0.008). A predictive algorithm that included these individual variables generated a clinical–germline risk score that successfully stratified patients into high- and low-risk groups with markedly different median PFS (low-risk: 18 months vs. high-risk: 7 months, log-rank P < 0.001), retaining discriminatory power across histological subgroups. Conclusions: The integration of specific germline variants as the output of advanced machine learning analysis of the exome with key clinical features provides accurate and novel predictive information for immunotherapy in NSCLC. This approach not only uncovers new genetic biomarkers but also supports the clinical adoption of composite risk scores for personalised precision immunotherapy, paving the way for improved patient selection and stratification. Non-small cell lung cancer Germline variation Immunotherapy biomarkers Machine learning Progression-free survival Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Background Non-small cell lung cancer (NSCLC) represents approximately 85% of all lung cancer cases and continues to be the leading cause of cancer-related mortality worldwide [ 1 ]. Immune checkpoint inhibitors (ICIs), notably those targeting the PD-1 and PD-L1 pathways, have transformed the treatment paradigms for advanced NSCLC and significantly extended survival in select patients [ 2 , 3 ]. However, only a subset of individuals achieves durable clinical responses, emphasising the need for robust biomarkers to guide optimal patient selection. Currently used tumour-based predictors, such as PD-L1 expression, tumour mutational burden, and microsatellite instability, show limited and inconsistent values across real-world populations [ 4 , 5 ]. Consequently, there is an urgent need to expand the biomarker repertoire beyond the tumour microenvironment. Recently, host-related factors, specifically germline genetic variations, have gained recognition as critical modulators of anti-tumour immune responses, influencing both treatment efficacy and immune-related toxicities [ 6 , 7 ]. Several retrospective and multi-omics studies have linked germline variants to immune phenotypes and clinical endpoints in ICI-treated NSCLC cohorts [ 8 – 10 ]. However, germline-informed predictive models have seldom been prospectively validated or integrated with clinical and tumour-derived data within real-world settings, limiting their clinical translation. Concurrently, advances in machine learning (ML) and artificial intelligence (AI) have enabled the analysis of complex multidimensional clinicogenomic datasets, uncovering subtle and non-linear predictors of therapeutic efficacy that escape traditional statistical methods[ 11 – 14 ]. Despite this technological progress, the translation of germline genomics into clinically actionable decision-making tools remains unfulfilled. Building on current evidence, we hypothesised that specific inherited variants are significantly associated with differential responses to ICIs in patients with advanced NSCLC and that machine learning-based integration of these signals facilitates robust risk stratification. Here, we present a prospective, multicentre study that systematically integrates whole-exome germline sequencing, comprehensive clinical data, and state-of-the-art ML algorithms to develop and validate a composite risk score for immunotherapy benefit in NSCLC, aiming to advance biomarker-driven precision medicine. Methods Study Design and Patient Cohort This prospective, multicentre study enrolled 117 adult patients (≥ 18 years) with histologically confirmed stage IIIB/IV NSCLC treated between January 2018 and June 2024 at two hospitals in Málaga (Hospital Universitario Virgen de la Victoria and Hospital Regional Universitario de Málaga), Spain. Eligible patients were scheduled to receive immune checkpoint inhibitor (ICI)-based therapy, either as monotherapy or in combination with chemotherapy. Key exclusion criteria included prior ICI exposure and a history of other malignancies (within five years, except for non-melanoma skin cancer). All patients provided written informed consent. The study was approved by the regional ethics committee and conducted in accordance with the Declaration of Helsinki. To ensure patient privacy during machine learning analysis, all clinical and genomic data were pseudonymized and stored on secure servers, in strict compliance with the General Data Protection Regulation (GDPR) (EU) 2016/679 Sample Collection and Clinical Data Peripheral blood samples were collected in EDTA tubes before the first ICI administration. Germline DNA was extracted using the QIAamp DNA Mini Kit (Qiagen) and quantified with Qubit dsDNA BR Assay Kit (Thermo Fisher Scientific). Comprehensive clinical data were collected from electronic medical records, including age, sex, histological subtype (adenocarcinoma [LUAD], squamous cell carcinoma [LUSC], or others), TNM stage (IIIB vs. IV), treatment regimen, PD-L1 expression status, metastatic sites, and ECOG performance status. Treatment efficacy was evaluated according to the RECIST v1.1 criteria [ 15 ], with tumour assessment performed every 8–12 weeks according to clinical practice. Clinical outcomes were categorised as Complete Response (CR), Partial Response (PR), Stable Disease (SD), or Progressive Disease (PD). [ 15 ] Whole-Exome Library Preparation and Sequencing Libraries were prepared from 50 ng of germline DNA using the Twist Human Core Exome 2.0 Kit (Twist Bioscience), which targets a total of 33 Mb of coding regions. The quality of the libraries was assessed using an Agilent 2100 Bioanalyzer. Sequencing was performed on an Illumina NovaSeq 6000 platform (2 × 150 bp), achieving a mean coverage of > 30x. FastQC v0.11.9 [ 16 ] and Fastp v0.39. [ 17 ] were used for raw data quality control and adapter trimming, respectively. Bioinformatics Processing and Variant Annotation Sequencing reads were aligned to the GRCh38 human reference genome using the BWA-MEM. Duplicates were marked, and base quality scores were recalibrated using Picard and GATK. Germline variants were called with GATK HaplotypeCaller, filtered per GATK best practices [ 18 ] (minimum depth > 30x, QD > 2.0, FS 40.0), and restricted to the 40–60% allelic fraction to minimise clonal haematopoietic artefacts. Functional annotation was performed using CRAVAT [ 19 ] and gnomAD v3.1.2 [ 20 ] for allele frequency analysis. Variants were classified according to pathogenicity, following The American College of Medical Genetics and Genomics (ACMG) recommendations [ 21 ]. Only rare (MAF < 1%), exonic, splice-site, and high-impact intronic variants were selected. To reduce data redundancy and minimise multicollinearity, we applied a pairwise correlation filter; for variant pairs with high correlation (Pearson’s r ≥ 0.70), only one representative variant was retained for downstream analysis. Variant annotation was blinded to the clinical outcomes. Statistical Analysis Associations between genetic variants and clinical benefits were assessed using Fisher’s exact test and logistic regression with FDR correction. To align with the concept of disease control, the outcome variable was dichotomised into responders (patients achieving CR, PR, or SD) and non-responders (patients with PD). Progression-free survival (PFS) and overall survival (OS) were compared using Kaplan–Meier curves and log-rank tests. Analyses were run in R v4.2.0 (packages: tidyverse, survival, xgboost). Statistical significance was set at P < 0.05. Machine Learning and Predictive Modelling To ensure robust phenotypic separation, we first defined a cohort of 'extreme responders' (non-responders 12 months). Using this cohort, we considered four modelling scenarios: (1) all selected germline variants, (2) exonic-only variants, (3) all variants per LUAD subgroup, and (4) exonic variants of LUAD only. For model construction, the dataset corresponding to these extreme phenotypes was split into training (70%) and test (30%) sets, and 1,000 iterations were performed to evaluate model stability and feature importance. Within each training set, hyperparameter tuning used stratified 5-fold cross-validation to minimise data leakage. To minimise overfitting and ensure rigorous evaluation, model construction was performed on a training set comprising 70% of the extreme responder population. For internal validation, we employed a temporal hold-out strategy, reserving the subset of extreme responder patients recruited during the final 12 months of the study period (n = 37). This temporal split mimics a prospective clinical scenario in which the model is tested on future patients, providing a more robust estimate of predictive performance than random splitting. The optimised model from each iteration was evaluated on the test set to assess predictive performance (AUC). The AUC values reflect the predictive accuracy of the immunotherapy response within the held-out test sets to ensure generalisability. Feature selection, model stability, and variable importance metrics were collated across the iterations. Subsequent survival modelling grouped patients according to genotype status (variant carriers vs. non-carriers) and compared outcomes using the log-rank test. To define the optimal thresholds, the surv_cutpoint function from the Survminer R package was used to identify the cutoff value that maximised the log-rank statistic. Based on the biological premise that mechanisms facilitating initial immune recognition (response) are essential for durable disease control, we hypothesised that robust response predictors would also stratify survival outcomes. Consequently, two composite risk scores were generated by integrating variables that demonstrated consistency across both the ML response model and univariate PFS analyses: (i) The Global Clinical-Germline Score (all patients), comprising six adverse predictors (five molecular, one clinical) and three favourable predictors (two molecular, one clinical), and (ii) The LUAD-Specific Score (adenocarcinoma only), comprising three adverse predictors and three favourable predictors. Adverse predictors (OR > 1) were incorporated as penalties by adding their log(OR) values, whereas favourable predictors (OR < 1) were treated as protective factors by subtracting log(OR) values. This weighting strategy reflects the magnitude of the association between each variable and the clinical outcome. The thresholds for the high- and low-risk groups were determined using an optimal cutoff based on (a) a minimum subgroup size of 10% and (b) maximisation of survival curve separation. To assess threshold robustness, a nonparametric bootstrap procedure (1,000 resamples) was used to estimate the 95% confidence interval of the optimal cutoff distribution. External Validation Cohorts To evaluate the robustness and generalisability of the derived clinical-germline risk score, we analysed two independent validation cohorts. First, to assess the performance of the gene signature in an external NSCLC population and given the scarcity of public datasets with matched germline-clinical immunotherapy data, we queried cBioPortal for Cancer Genomics. We selected the dataset from Rizvi et al. 2015/Hellmann et al. 2018, comprising 75 patients treated with immune checkpoint inhibitors [ 22 ]. As this dataset primarily provides somatic sequencing data, validation was restricted to assessing the prognostic value of the identified gene signature at the gene level. Second, to explore the utility of the score across different tumour types (cross-tumour validation), we included an independent in-house cohort of 19 patients with metastatic melanoma. These patients were treated with anti-PD-1/PD-L1 monotherapy at Hospital Regional de Malaga, following the same inclusion criteria and ethical standards as those of the main cohort. Germline sequencing and clinical data collection for this subset followed the pipeline described above. Data and Code Availability Raw sequencing data, genotype matrices, and scripts can be obtained from the corresponding authors upon reasonable request and institutional approval. Results Patients’ characteristics A total of 117 patients with advanced NSCLC were enrolled. The median age was 67.5 years (range: 46–85 years), and 74.4% were male. Histologically, lung adenocarcinoma (LUAD) accounted for 59.8% of the cases, while 30.8% were squamous cell carcinomas (LUSC). Although 19.1% of patients were initially diagnosed with locally advanced disease (Stage IIIB), all presented with metastatic progression or Stage IV disease at the time of ICI initiation. Among those with metastatic spread, the most frequent sites were the lungs (51.3%) and the lymph nodes (47.9%). ICI monotherapy was administered in 25.6% of cases, with combination chemo-ICI or other regimens administered in 74.4% of cases. At the first evaluation per RECIST v1.1, the Objective Response Rate (CR or PR) was 17.1% in evaluable patients, whereas 65% achieved Disease Control (CR, PR, or SD) by 3 months. The detailed cohort characteristics are presented in Table 1 . Table 1 Demographic and Clinical Features of the Study Population Summary of the baseline characteristics of the 117 patients with advanced NSCLC included in the study. Variable Category n (117) % Sex Female (F) 30 25.6 Male (M) 87 74.4 Histology LUAD 70 59.8 LUSC 36 30.8 NSCLC NS 7 6.0 Other 4 3.4 Stage at diagnosis IIIB (locally advanced) 22 19.13 IV (Metastatic) 83 72.17 PD-L1 expression 50% 10 8.5 Not evaluated 9 7.7 Therapy type ICI monotherapy 30 25.6 CT + ICI 43 36.8 CT + ICI + Others 39 33.3 Immunotherapy combo 1 0.9 RECIST 1st evaluation (8–12 weeks) Complete Response (CR) 5 4.3 Partial Response (PR) 15 12.8 Stable Disease (SD) 3 2.6 Progressive Disease (PD) 11 9.4 Not evaluable 83 70.9 Clinical Benefit At 3 months (Responder vs Non-Responders) Responders 76 65.0 Non-responders 41 35.0 Clinical Benefit At 12 months (Responder vs Non-Responders) Responders 39 33.3 Non-responders 69 59.0 M1 Metastasis Brain 14 12.0 Liver 12 10,25 Bone 40 34,18 Lung 60 51,28 Skin 30 25,64 Lymph node 56 47,86 * Abbreviations: CT, chemotherapy; ICI, immune checkpoint inhibitor; IT, immunotherapy; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma; NSCLC NOS, non-small cell lung cancer not otherwise specified; NE, not evaluable; M1, distant metastasis. ¨Other" histology includes large cell carcinoma and adenosquamous carcinoma. M1 sites refer to distant metastases confirmed at the baseline. Clinical Variables associate with Immunotherapy Response Multivariable logistic regression confirmed that lymph node metastasis was the sole independent clinical predictor of poor response to ICIs (OR 7.9, 95% CI 1.8–43.3, P = 0.008). Given that all patients were treated in the advanced/metastatic setting, the tumour stage was not included as a covariate. Regarding the metastatic burden, no specific organ involvement (brain, liver, bone, lung, or skin) other than lymph nodes was significantly associated with the outcome, as shown in Fig. 1 . Neither treatment modality (ICI monotherapy vs. combination), sex, PD-L1 expression strata, nor ECOG performance status were significantly associated with clinical benefits. Response Germline Variant Discovery and Model Performance To identify germline predictors of immunotherapy response, we implemented a comprehensive variant filtering and prioritisation pipeline (Fig. 2 a). From an initial pool of 108,944 called variants, strict quality control and linkage disequilibrium pruning (Pearson’s r < 0.70) yielded a refined set of candidates. Univariate statistical filtering further narrowed this pool to those showing the strongest association with clinical benefit. The predictive performance of the machine learning models was evaluated across four analytical scenarios. The XGBoost model utilising all exonic variants achieved the highest accuracy, with a cross-validated median area under the ROC curve (AUC) of 0.845 (IQR 0.742–0.879) in the held-out test set (Fig. 2 b). Comparative modelling consistently demonstrated the superiority of XGBoost over LASSO regression across all datasets. For the LUAD-specific cohort, XGBoost achieved an AUC of 0.82, compared to 0.56 for LASSO. Similarly, in models restricted to exonic variants within LUAD, XGBoost maintained a robust performance (AUC 0.77) compared to LASSO (AUC 0.72). Feature importance analysis revealed distinct genetic signatures depending on the histological context. In the global analysis (all patients), the most recurrently selected loci (> 70% of iterations) included variants in OR51M1, C22orf34, DENND2B, SLC6A16, ZNF148, SIGLEC11, ZNF454 , and PDE4E genes. Conversely, the analysis focused specifically on the LUAD subgroup highlighted distinct drivers, identifying ARAP3, CCZ1B, ARHGAP22 , and OR10H5 as significant contributors. The frequent selection of these variants suggests their potential role in germline-mediated modulation of the immune response. Statistical Validation and Prognostic Characterization of Prioritised Variants To individually validate the germline features prioritised by the best-performing XGBoost model (exonic variants) and quantify their effect sizes, we performed univariate logistic regression analysis. This post-hoc analysis confirmed that several germline variants were significantly associated (P T ( SLC6A16 ), chr19:49960436_>T ( SIGLEC11 ), chr5:178964901 A > C ( ZNF454 ), chr5:60460204 A > G ( PDE4E ), chr10:48451594 T > C ( ARHGAP22 ), and chr19:15794719 C > A ( OR10H5 ) were associated with a higher likelihood of non-response (resistance). In contrast, chr11:8766678 A > C ( DENND2B ), chr11:118643907 C > T ( PHLDB1 ), chr5:141671669 C > T ( ARAP3 ), and chr7:6822249 G > C ( CCZ1B ) were associated with a positive treatment outcome (Table 2 ). Table 2 Prioritised Germline Variants and Statistical Associations with Immunotherapy Outcomes List of the top germline variants selected by machine learning models and univariate logistic regression, showing odds ratios, confidence intervals, and p-values for immunotherapy response, progression-free survival (PFS), and overall survival (OS). Genes, variant positions, variant types (synonymous, non-synonymous, and intronic), and relevant molecular functions were annotated for each locus Univariante PFS OS Gene Variant (chr:position/ref/alt) Variant Type OR 95% CI p-val Log-Rank P-val Log-Rank P-val OR51M1 chr11:5390165:C > A Non-synonymous 2.94 (0.7–14.5) 0.1 0.89 0.96 DENND2B chr11:8766678:A > C Intronic 0.28 (0.08 -0,86) 0.03* 0.067 0.22 SLC6A16 chr19:49309146:G > T Intronic 4.90 (1,73- 15.8) 0.004** 0.0048** 0.0057** SIGLEC11 chr19:49960436:_:T Intronic 8.76 (2.5–41) 0.002** 0.03* 0.00021** C22orf34 chr22:49624397:G > A Intronic 3.74 (0.79- 26.) 0.1 0.15 0.043* ZNF148 chr3:125232977:C > T Non-synonymous 2.14 (0,58 − 8.9) 0.3 0.96 0.84 ZNF454 chr5:178964901:A > C Non-synonymous 4.71 (1.28–22.6) 0.03* 0.34 0.17 PDE4E chr5:60460204:A > G Intronic 6.13 (2.5–21.1) 0.002** 0.0051** 0.0001** PHLDB1 chr11:118643907:C > T Non-synonymous 0.15 (0.04-0,44) 0.001** 0.036* 0.011* ARAP3 chr5:141671669:C > T Non-synonymous 0.22 (0,06- -0,63) 0.007** 0.085 0.0001** ARHGAP22 chr10:48451594:T > C Intronic 3.64 (1.32–10.5) 0.01** 0.049* 0.05* CCZ1B chr7:6822249:G > C Intronic 0.14 (0.04–0.42) 0.001** 0.0016** 0.00031** OR10H5 chr19:15794719:C > A Non-synonymous 7.03 (2.03–33.1) 0.005** 0.0075** 0.028* Furthermore, we assessed progression-free survival (PFS) to corroborate these findings. The median PFS was compared between patients harbouring the variant (variant-positive) and those with the reference genotype (wild type). Consistent with the logistic regression results, the CCZ1B variant (chr7:6822249 G > C) was associated with significantly extended PFS (21 vs. 9 months; P = 0.0016), aligning with its protective odds ratio OR (0.14). A similar survival benefit was observed for PHLDB1 (chr11:118643907 C > T; 12 vs. 8 months; P = 0.036). Conversely, the presence of variants in SLC6A16 (chr19:49309146 G > T ) (P = 0.0048), OR10H5( chr19:15794719 C > A) (P = 0.0075), PDE4E (chr5:60460204 A > G ) (P = 0.0051), ARHGAP22 (chr10:48451594 T > C) (P = 0.049), and SIGLEC11 (chr19:49960436_>T )(P = 0.03) was associated with significantly reduced PFS compared to non-carriers. Variants in DENND2B (chr11:8766678 A > C ) and ARAP3 (chr5:141671669 C > T) showed a trend toward PFS benefits (P = 0.067 and P = 0.085, respectively), although the differences were not significant. Finally, we evaluated the Overall Survival (OS). Most notably, the CCZ1B (chr7:6822249 G > C ) variant conferred a substantial survival advantage (median OS, 22 vs. 7 months; P T and ARAP3 (chr5:141671669 C > T) variants achieved significance for OS (despite missing the PFS threshold), variants in PDE4E (chr5:60460204 A > G), SLC6A16 (chr19:49309146 G > T), SIGLEC11 ( chr19:49960436_>T), ARHGAP22 ( chr10:48451594 T > C), and OR10H5 (chr19:15794719 C > A ) were associated with a marked decline in OS. Interestingly, the C22orf34 variant (chr22:49624397 G > A), which was not significant in the response or PFS analyses, was identified as a novel predictor of poor OS (P = 0.043). The representative Kaplan–Meier curves are shown in Fig. 4 . Multivariable Analysis of Prioritized Germline and Clinical Data To determine the independent predictive value of the identified biomarkers, we conducted a comprehensive multivariable logistic regression analysis. This model integrated the germline variants selected by the XGBoost algorithm, specifically those showing consistency in the univariate response and progression-free survival (PFS) assessments, along with key clinical covariates (sex, histology, PD-L1 expression, treatment type, and metastatic sites). In all model iterations, lymph node metastasis consistently emerged as a robust independent predictor of resistance, yielding Odds Ratios (OR) for non-response ranging from approximately 5.9 to 21.0 (P < 0.05) depending on the variant combination (Fig. 5 ). Regarding germline features, six variants maintained a significant correlation with treatment outcomes after adjusting for clinical factors. Variants in SLC6A16 (chr19:49309146 G > T), PDE4E (chr5:60460204 A > G), and ARHGAP22 (chr10:48451594 T > C) were confirmed as independent predictors of poor response. Patients carrying these variants experienced a significantly higher likelihood of resistance and shorter median PFS. Conversely, the PHLDB1 (chr11:118643907C > T), CCZ1B (chr7:6822249G > C), and OR10H5 (chr19:15794719C > A) variants were independently associated with favourable outcomes, exhibiting protective odds ratios in the regression model, and correlated with extended median PFS in carriers (Fig. 5 ). No consistent independent associations were found for other clinical factors, including distant metastases, immunotherapy regimen, sex, PD-L1 expression, or ECOG status. Consequently, lymph node involvement and the identified germline profile were the most reliable independent indicators of the clinical benefit. Notably, these predictive effects were observed for both non-synonymous and intronic variants, suggesting diverse functional mechanisms of immune modulation. Development and Validation of the Clinical-Germline Prognostic Score Finally, to translate these findings into a clinical tool, we constructed a composite risk score by integrating the prioritised germline variants with lymph node status, weighting each variable by its corresponding regression coefficient (log-odds ratio). Patients were stratified into high- and low-risk groups based on the optimal cutoff determined by the surv_cutpoint algorithm. Kaplan–Meier analysis demonstrated a robust distinction in clinical outcomes between the strata. In the overall cohort, individuals in the low-risk category experienced significantly longer progression-free survival (PFS) than those in the high-risk group (median 18.0 vs. 7.0 months; log-rank P < 0.001) (Fig. 6 a). This stratification power was maintained in the LUAD subgroup (median PFS 19.0 vs. 5.0 months; P = 0.012) (Fig. 6 b). Validation in Independent Cohorts To assess the robustness of the score, we first applied a temporal hold-out strategy and tested the model on a subset of patients recruited during the final study period (n = 22). Using the predefined cutoff, the score successfully stratified PFS (P = 0.002; median 18 vs. 4 months), confirming internal stability (Fig. 7 a-b). Subsequently, we explored the score's utility in external settings using an independent NSCLC cohort (cBioPortal[ 22 ] ) and an in-house melanoma cohort treated with ICIs. Since only somatic sequencing data were available for the external NSCLC cohort, we evaluated the gene-level impact of the signature. Although survival differences did not reach statistical significance, likely due to biological distinctions between somatic and germline alterations, similar stratification trends were observed, with low-risk patients consistently showing superior survival curves. This suggests that the identified gene pathways may play a conserved role in modulating immunotherapy outcomes across contexts (Fig. 7 c-d). Discussion Our study provides new evidence that germline variants, when analysed using advanced machine learning models, deliver valuable predictive information for immunotherapy response and outcomes in non-small cell lung cancer (NSCLC) [ 23 ]. Crucially, the robustness of these findings was supported by rigorous internal temporal validation and further explored in independent external cohorts, reinforcing the potential generalisability of the identified biomarkers. Unlike tumour-restricted or PD-L1-based biomarkers, inherited genomic variations appear to play an active role in determining therapeutic response, reflecting constitutional influences on host immune competence and tumour-immune interactions. Variants within genes such as CCZ1B, SIGLEC11, PDE4E, ARHGAP22, PHLDB1, SLC6A16 , and OR10H5 were consistently associated with progression-free survival (PFS), supporting the notion that the germline genome encodes mechanisms that modulate immune activation or resistance. Through an agnostic approach, the machine learning algorithm identified several genes associated with immunotherapy response, some of which are linked to tumourigenic processes or have recently been implicated in pathways potentially regulating immunotherapy response in cancer. For example, SIGLEC11 is a membrane receptor that negatively regulates the immune response by recognising and binding to sialoglycans on tumour cells, facilitating immune evasion and metastasis via the suppression of anti-tumour immunity [ 24 ]. Other relevant genes include PDE4E , a cAMP-specific phosphodiesterase linked to immunotherapy efficacy in LUAD [ 25 ], as well as SLC6A16 , whose elevated expression in tumour cells has been shown to promote immunotherapy resistance by impairing CD8⁺ T cell function and reducing immune infiltration [ 26 ]. Mechanistic and Prognostic Implications Beyond these specific gene functions, these findings collectively emphasise the utility of genome-wide, hypothesis-free approaches to uncover germline-mediated immunotherapy response mechanisms. However, other genes identified by this model have not yet been associated with tumour development or immune-related pathways. These findings may reflect statistical noise or the involvement of previously unrecognised germline mechanisms, warranting further functional investigations [ 27 ]. Among the favourable prognostic markers, CCZ1B showed the greatest survival benefit. The CCZ1B gene is involved in vesicular trafficking [ 28 ]; however, its specific role in boosting the cancer immune response remains to be elucidated. In contrast, variants in OR10H5 were associated with resistance. Although OR10H5 is canonically involved in olfaction, the identification of variants within this highly polymorphic gene family warrants caution because of the potential for mapping artefacts. While our stringent quality filtering minimises technical noise, we cannot exclude the possibility that this variant functions as a surrogate marker in a linkage disequilibrium with adjacent immune-regulatory loci. Future functional studies are essential to distinguish between the direct immunomodulatory role of OR10H5 and the passenger effect. Notably, however, its expression has previously been implicated in cancer biology; for instance, OR10H5 is part of a 19-gene prognostic signature in colorectal cancer, which correlates with improved patient survival [ 29 ]. Our data suggest a broader, context-dependent role in oncogenesis that warrants further investigation in NSCLC Additionally, the C22orf34 gene (a long non-coding RNA) was significantly associated with overall survival (OS) but not with PFS, suggesting that its prognostic relevance may be limited to long-term outcomes. Notably, C22orf34 has been previously associated with an increased risk of drug-induced interstitial lung disease (DIILD) in cancer patients [ 27 ], highlighting a potential link to treatment-related toxicity. Beyond single-variant effects, the integrated germline–clinical score efficiently stratified patients according to PFS. The persistence of lymph node metastasis as an independent adverse factor underscores the multifactorial nature of resistance. Methodologically, the superior performance of XGBoost (AUC 0.845) over classical penalised regression approaches like LASSO (AUC 0.56–0.72) demonstrates the critical advantage of ensemble learning in capturing non-linear genotype–phenotype relationships. While traditional linear models often fail to account for the complex epistatic interactions inherent in genomic data, our ML approach effectively modelled these dependencies. This performance compares favourably with reported predictive metrics for standard biomarkers, such as PD-L1 alone, in real-world settings, suggesting that germline features capture a component of "missing heritability" not addressed by current tumour-only markers. Notably, predictive models using exomic data yielded slightly superior AUC compared to the entire variant set. This highlights the importance of feature selection, even when exonic variants directly encode protein-altering changes. The reduction of variables to the most informative subset improved model accuracy by excluding redundant or noisy features and minimising overfitting, a finding consistent with recent studies applying hybrid dimension reduction techniques to cancer genomics [ 30 – 32 ]. Nonetheless, this study had limitations. The single-country design and moderate sample size may limit generalisability and affect the effect estimates for rare variants. Despite strong internal validation and exploratory external validation cohorts, strictly matched germline replication studies in multi-ethnic NSCLC populations are needed to confirm these findings. A further limitation of our external validation is the reliance on somatic sequencing data. Standard somatic variant calling pipelines often filter out germline signals as 'normal' background; therefore, our analysis likely captured somatic hits in the same genes rather than the original germline variants. However, the fact that somatic alterations in these loci also stratify risk supports the hypothesis that these genes are functional bottlenecks in tumour-immune interactions, regardless of the genomic origin (germline vs. somatic) of the alteration. Additionally, functional studies are required to causally connect these germline variations with immune-modulatory mechanisms. Conclusion Our findings establish a foundation for a comprehensive approach to discover immunotherapy biomarkers and stratify risk by integrating clinical and germline genetic data. This study illustrates how advanced AI-driven modelling can harness the predictive capabilities of inherited genetic variations, offering a pathway toward more personalised and precise immunotherapy strategies for lung cancer. Abbreviations NSCLC Non-small cell lung cancer LUAD Lung adenocarcinoma LUSC Lung squamous cell carcinoma ICI Immune checkpoint inhibitor PD-1 Programmed cell death protein 1 PD-L1 Programmed death-ligand 1 TMB Tumor mutational burden PFS Progression-free survival OS Overall survival CR Complete response PR Partial response SD Stable disease PD Progressive disease OR Odds ratio CI Confidence Interval DNA Deoxyribonucleic Acid RNA Ribonucleic Acid NGS Next-generation sequencing WES Whole-exome sequencing ML Machine learning AI Artificial intelligence EDTA Ethylenediaminetetraacetic acid ECOG Eastern Cooperative Oncology Group RECIST Response Evaluation Criteria in Solid Tumours QD Quality by depth VAF Variant allele frequency ACMG American College of Medical Genetics and Genomics FDR False discovery rate ROC Receiver operating characteristic AUC Area under the curve cAMP Cyclic adenosine monophosphate Declarations Ethics approval and consent to participate This study was conducted in accordance with the Declaration of Helsinki and approved by the Provincial Research Ethics Committee of Málaga (PEIBA), reference number 26-10-2017. Written informed consent was obtained from all participants included in the study. Consent for publication All authors have provided consent for publication. Availability of data and materials The datasets generated and/or analysed during the current study are available from the corresponding author upon reasonable request. Competing interests The authors declare no conflict of interest. Funding This work was supported by Consejería de Conocimiento, Investigación y Universidad, Junta de Andalucía under grant number PAIDI P21-01002. Funding: JO and IB holds ‘Nicolas Monardes’ research contracts from the Andalusian Regional Ministry Health (JO: C1-0003-2023) Authors’ contributions JO, AR, IB: conceptualization , AR , JLO, MRG, GPL: methodology and software, EPR, JCB , JZ, AC: clinical data curation, AGH, MGB, LCFO, BMG: experimental procedures, formal analysis, JO, AR: writing-original draft preparation, JO, AGR, AR: writing-review & editing, JO, ARD: supervision. All authors have read and approved the final manuscript. Acknowledgements We thank the Picasso Supercomputer, University of Málaga (UMA), for providing the high-performance computing resources essential for performing the genomic and machine learning analyses presented in this study. References Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71:209–49. https://doi.org/10.3322/caac.21660 . Reck M, Rodríguez-Abreu D, Robinson AG, Hui R, Csőszi T, Fülöp A, et al. Pembrolizumab versus Chemotherapy for PD-L1–Positive Non–Small-Cell Lung Cancer. New Engl J Med Mass Med Soc. 2016;375:1823–33. https://doi.org/10.1056/NEJMoa1606774 . Gandhi L, Rodríguez-Abreu D, Gadgeel S, Esteban E, Felip E, De Angelis F, et al. Pembrolizumab plus Chemotherapy in Metastatic Non-Small-Cell Lung Cancer. N Engl J Med. 2018;378:2078–92. https://doi.org/10.1056/NEJMoa1801005 . Wallenta Law J, Bapat B, Sweetnam C, Mohammed H, McBratney A, Izano MA, et al. Real-World Impact of Comprehensive Genomic Profiling on Biomarker Detection, Receipt of Therapy, and Clinical Outcomes in Advanced Non-Small Cell Lung Cancer. JCO Precis Oncol. 2024;8:e2400075. https://doi.org/10.1200/PO.24.00075 . Hellmann MD, Ciuleanu T-E, Pluzanski A, Lee JS, Otterson GA, Audigier-Valette C, et al. Nivolumab plus Ipilimumab in Lung Cancer with a High Tumor Mutational Burden. N Engl J Med. 2018;378:2093–104. https://doi.org/10.1056/NEJMoa1801946 . Cao W, Tang Q, Zeng J, Jin X, Zu L, Xu S. A Review of Biomarkers and Their Clinical Impact in Resected Early-Stage Non-Small-Cell Lung Cancer. Cancers (Basel). 2023;15:4561. https://doi.org/10.3390/cancers15184561 . Middha P, Thummalapalli R, Quandt Z, Balaratnam K, Cardenas E, Falcon CJ, et al. Germline prediction of immune checkpoint inhibitor discontinuation for immune-related adverse events. J Immunother Cancer. 2025;13:e011273. https://doi.org/10.1136/jitc-2024-011273 . Chin IS, Khan A, Olsson-Brown A, Papa S, Middleton G, Palles C. Germline genetic variation and predicting immune checkpoint inhibitor induced toxicity. npj Genom Med Nat Publishing Group. 2022;7:1–14. https://doi.org/10.1038/s41525-022-00345-6 . Gusev A. Germline mechanisms of immunotherapy toxicities in the GWAS era. Immunol Rev. 2023;318:138–56. https://doi.org/10.1111/imr.13253 . Weidhaas J, Marco N, Scheffler AW, Kalbasi A, Wilenius K, Rietdorf E, et al. Germline biomarkers predict toxicity to anti-PD1/PDL1 checkpoint therapy. J Immunother Cancer. 2022;10:e003625. https://doi.org/10.1136/jitc-2021-003625 . Monaco A, Pantaleo E, Amoroso N, Lacalamita A, Lo Giudice C, Fonzino A, et al. A primer on machine learning techniques for genomic applications. Comput Struct Biotechnol J. 2021;19:4345–59. https://doi.org/10.1016/j.csbj.2021.07.021 . Al-Tashi Q, Saad MB, Muneer A, Qureshi R, Mirjalili S, Sheshadri A, et al. Machine Learning Models for the Identification of Prognostic and Predictive Cancer Biomarkers: A Systematic Review. Int J Mol Sci. 2023;24:7781. https://doi.org/10.3390/ijms24097781 . Bhushan R, Rai S, Jangra P, Sonker P, Singh N, Lakshami R, et al. The role of artificial intelligence in predicting cancer immunotherapy response. Oral Oncol Rep. 2025;16:100767. https://doi.org/10.1016/j.oor.2025.100767 . Rakaee M, Tafavvoghi M, Ricciuti B, Alessi JV, Cortellini A, Citarella F, et al. Deep Learning Model for Predicting Immunotherapy Response in Advanced Non-Small Cell Lung Cancer. JAMA Oncol. 2025;11:109–18. https://doi.org/10.1001/jamaoncol.2024.5356 . Eisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, Sargent D, Ford R, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer. 2009;45:228–47. https://doi.org/10.1016/j.ejca.2008.10.026 . Van der Auwera GA, Carneiro MO, Hartl C, Poplin R, del Angel G, Levy-Moonshine A, et al. From FastQ data to high confidence variant calls: the Genome Analysis Toolkit best practices pipeline. Curr Protoc Bioinf. 2013;11:11. https://doi.org/10.1002/0471250953.bi1110s43 . Chen S, Zhou Y, Chen Y, Gu J. fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics. 2018;34:i884–90. https://doi.org/10.1093/bioinformatics/bty560 . McKenna A, Hanna M, Banks E, Sivachenko A, Cibulskis K, Kernytsky A, et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 2010;20:1297–303. https://doi.org/10.1101/gr.107524.110 . Masica DL, Douville C, Tokheim C, Bhattacharya R, Kim R, Moad K, et al. CRAVAT 4: Cancer-Related Analysis of Variants Toolkit. Cancer Res. 2017;77:e35–8. https://doi.org/10.1158/0008-5472.CAN-17-0338 . Karczewski KJ, Francioli LC, Tiao G, Cummings BB, Alföldi J, Wang Q, et al. The mutational constraint spectrum quantified from variation in 141,456 humans. Nat Nat Publishing Group. 2020;581:434–43. https://doi.org/10.1038/s41586-020-2308-7 . Richards S, Aziz N, Bale S, Bick D, Das S, Gastier-Foster J, et al. Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. Genet Med Nat Publishing Group. 2015;17:405–23. https://doi.org/10.1038/gim.2015.30 . Gao J, Aksoy BA, Dogrusoz U, Dresdner G, Gross B, Sumer SO, et al. Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal. Sci Signal. 2013;6:pl1. https://doi.org/10.1126/scisignal.2004088 . Chao BN, Carrick DM, Filipski KK, Nelson SA. Overview of Research on Germline Genetic Variation in Immune Genes and Cancer Outcomes. Cancer Epidemiol Biomarkers Prev. 2022;31:495–506. https://doi.org/10.1158/1055-9965.EPI-21-0583 . Yin J, Lu Y, Liu Y, Shi Q, Shi M, Zhu Z, et al. SIGLEC11 promotes M2 macrophage polarization through AKT-mTOR signaling and facilitates the progression of gastric cancer. J Immunother Cancer. 2025;13:e010162. https://doi.org/10.1136/jitc-2024-010162 . Feng B, Pan B, Huang J, Du Y, Wang X, Wu J, et al. PDE4D/cAMP/IL-23 axis determines the immunotherapy efficacy of lung adenocarcinoma via activating the IL-9 autocrine loop of cytotoxic T lymphocytes. Cancer Lett. 2023;565:216224. https://doi.org/10.1016/j.canlet.2023.216224 . Ren H-L, Zhang S-H, Li P-Y. The multifaceted role of phosphodiesterase 4 in tumor: from tumorigenesis to immunotherapy. Front Immunol. 2025;16:1528932. https://doi.org/10.3389/fimmu.2025.1528932 . Udagawa C, Horinouchi H, Shiraishi K, Kohno T, Okusaka T, Ueno H, et al. Whole genome sequencing to identify predictive markers for the risk of drug-induced interstitial lung disease. PLoS ONE. 2019;14:e0223371. https://doi.org/10.1371/journal.pone.0223371 . Wright PK. Targeting vesicle trafficking: an important approach to cancer chemotherapy. Recent Pat Anticancer Drug Discov. 2008;3:137–47. https://doi.org/10.2174/157489208784638730 . Abdul Aziz NA, Mokhtar NM, Harun R, Mollah MMH, Mohamed Rose I, Sagap I, et al. A 19-Gene expression signature as a predictor of survival in colorectal cancer. BMC Med Genomics. 2016;9:58. https://doi.org/10.1186/s12920-016-0218-1 . Zengin HY, Karabulut E. Biomarker detection using corrected degree of domesticity in hybrid social network feature selection for improving classifier performance. BMC Bioinformatics. 2023;24:407. https://doi.org/10.1186/s12859-023-05540-5 . Wang Z, Zhou Y, Takagi T, Song J, Tian Y-S, Shibuya T. Genetic algorithm-based feature selection with manifold learning for cancer classification using microarray data. BMC Bioinformatics. 2023;24:139. https://doi.org/10.1186/s12859-023-05267-3 . Hauskrecht M, Pelikan R, Valko M, Lyons-Weiler J. Feature Selection and Dimensionality Reduction in Genomics and Proteomics. In: Werner Dubitzky MG, Berrar D, editors. Fundamentals of Data Mining in Genomics and Proteomics. Springer; 2006. pp. 149–72. https://doi.org/10.1007/978-0-387-47509-7 . Supplementary Files GAxgboost.png SupplementaryMaterialtable1.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 10 Feb, 2026 Reviewers invited by journal 16 Jan, 2026 Editor assigned by journal 13 Jan, 2026 First submitted to journal 07 Jan, 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-8455214","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":575380057,"identity":"e43a2b34-a3e0-4515-947d-9bc62bb88734","order_by":0,"name":"Andrea González-Hernández","email":"","orcid":"","institution":"centro de investigación y terapias avanzadas del cáncer","correspondingAuthor":false,"prefix":"","firstName":"Andrea","middleName":"","lastName":"González-Hernández","suffix":""},{"id":575380058,"identity":"eee006e0-6d82-449d-a87f-2bf5cec787f9","order_by":1,"name":"Alberto Ríos","email":"","orcid":"","institution":"centro de investigación y terapias avanzadas del cáncer","correspondingAuthor":false,"prefix":"","firstName":"Alberto","middleName":"","lastName":"Ríos","suffix":""},{"id":575380059,"identity":"50a104f9-6f13-47c3-95ac-efb404e9189e","order_by":2,"name":"Juan Luis Onieva","email":"","orcid":"","institution":"IBIMA","correspondingAuthor":false,"prefix":"","firstName":"Juan","middleName":"Luis","lastName":"Onieva","suffix":""},{"id":575380060,"identity":"4831884b-3368-4229-ab4a-9128baf3475d","order_by":3,"name":"Alexandra Cantero","email":"","orcid":"","institution":"Hospital Regional Universitario de Málaga: Hospital Regional Universitario de Malaga","correspondingAuthor":false,"prefix":"","firstName":"Alexandra","middleName":"","lastName":"Cantero","suffix":""},{"id":575380061,"identity":"01e8a89d-3652-4bd3-ba97-b0f5c482515c","order_by":4,"name":"Marina Rivero-Aguilar","email":"","orcid":"","institution":"IBIMA","correspondingAuthor":false,"prefix":"","firstName":"Marina","middleName":"","lastName":"Rivero-Aguilar","suffix":""},{"id":575380062,"identity":"56c5a673-9528-45c0-b51b-af675fb0b844","order_by":5,"name":"Guillermo Paz-López","email":"","orcid":"","institution":"IBIMA","correspondingAuthor":false,"prefix":"","firstName":"Guillermo","middleName":"","lastName":"Paz-López","suffix":""},{"id":575380063,"identity":"65b0368d-af28-4855-89ae-20ed42673935","order_by":6,"name":"Antonio Rueda-Dominguez","email":"","orcid":"","institution":"IBIMA","correspondingAuthor":false,"prefix":"","firstName":"Antonio","middleName":"","lastName":"Rueda-Dominguez","suffix":""},{"id":575380064,"identity":"39f4dc9d-937e-4b0b-b8b8-a9aa89a9c41f","order_by":7,"name":"María Garrido-Barros","email":"","orcid":"","institution":"IBIMA","correspondingAuthor":false,"prefix":"","firstName":"María","middleName":"","lastName":"Garrido-Barros","suffix":""},{"id":575380065,"identity":"80f9975e-3512-4ed8-af68-d72112c52da2","order_by":8,"name":"Beatriz Martínez-Gálvez","email":"","orcid":"","institution":"IBIMA","correspondingAuthor":false,"prefix":"","firstName":"Beatriz","middleName":"","lastName":"Martínez-Gálvez","suffix":""},{"id":575380066,"identity":"b87194a7-c15d-45fa-badf-efffb2c4b19b","order_by":9,"name":"Juan Zafra","email":"","orcid":"","institution":"IBIMA","correspondingAuthor":false,"prefix":"","firstName":"Juan","middleName":"","lastName":"Zafra","suffix":""},{"id":575380067,"identity":"4706aca5-7e7e-4583-80e8-360a31dde8c0","order_by":10,"name":"Laura Cristina Figueroa-Ortiz","email":"","orcid":"","institution":"Universidad de Malaga","correspondingAuthor":false,"prefix":"","firstName":"Laura","middleName":"Cristina","lastName":"Figueroa-Ortiz","suffix":""},{"id":575380068,"identity":"8208984a-5a86-429c-85f6-9aedbdea5f64","order_by":11,"name":"Elisabeth Pérez-Ruiz","email":"","orcid":"","institution":"IBIMA","correspondingAuthor":false,"prefix":"","firstName":"Elisabeth","middleName":"","lastName":"Pérez-Ruiz","suffix":""},{"id":575380069,"identity":"e9ffec44-e21b-41a5-adc1-e82043af1488","order_by":12,"name":"José Carlos Benitez","email":"","orcid":"","institution":"Hospital Universitario Virgen de la Victoria","correspondingAuthor":false,"prefix":"","firstName":"José","middleName":"Carlos","lastName":"Benitez","suffix":""},{"id":575380070,"identity":"10eb9a8a-eb29-4558-8528-7bb7a67eb864","order_by":13,"name":"Isabel Barragan","email":"","orcid":"","institution":"IBIMA","correspondingAuthor":false,"prefix":"","firstName":"Isabel","middleName":"","lastName":"Barragan","suffix":""},{"id":575380071,"identity":"71cfaa4b-da5e-49e4-9ccd-0be4b7315f0e","order_by":14,"name":"Javier Oliver","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAtklEQVRIiWNgGAWjYBACxgYGhgMMDDYgZuMBUrSkgZnEaYGCw2CSOC3M7WcMD3zMOW+3tv0w0JYam2jCDuvJMTg4c9vt5G1nEoFajqXlNhDU0pCWcJgXqMXsAFALY8NhIrT0PwNpOZdsdv4hsVpmJB8AajlgZ3aDaFtmPD4A9EtygtkNoC0JxPjFsD+x+cPHbXb2ZufTHz74UGNDhBaoikQwnUBIOQjIQ2l7YhSPglEwCkbBCAUAqElNMe89GyEAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0002-8073-4711","institution":"IBIMA","correspondingAuthor":true,"prefix":"","firstName":"Javier","middleName":"","lastName":"Oliver","suffix":""}],"badges":[],"createdAt":"2025-12-26 12:23:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8455214/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8455214/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":100676599,"identity":"f5330be7-47af-4b9e-9711-32a816601a61","added_by":"auto","created_at":"2026-01-20 11:21:24","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":18193,"visible":true,"origin":"","legend":"","description":"","filename":"Table1.docx","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/018c2ecc801c2d6ab7357ef7.docx"},{"id":100676357,"identity":"c080afa7-be2b-4646-9c7a-232e449b581f","added_by":"auto","created_at":"2026-01-20 11:19:18","extension":"docx","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":17998,"visible":true,"origin":"","legend":"","description":"","filename":"Table2.docx","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/f38b10275ac5ab361290b0ab.docx"},{"id":100676593,"identity":"50b6995d-3e7f-4545-a52e-f5c31a742030","added_by":"auto","created_at":"2026-01-20 11:21:16","extension":"xml","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19850,"visible":true,"origin":"","legend":"","description":"","filename":"jtrmJTRMD2523156.xml","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/e142a966b50c6f2e520e3ceb.xml"},{"id":100676596,"identity":"e27584d5-b10c-4993-af06-6ef945173ba9","added_by":"auto","created_at":"2026-01-20 11:21:20","extension":"xml","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1443,"visible":true,"origin":"","legend":"","description":"","filename":"JTRMD2523156167176.go.xml","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/6ccba2e838466624a098dd49.xml"},{"id":100676314,"identity":"b63cf541-e233-4525-8bd7-f26757b1012d","added_by":"auto","created_at":"2026-01-20 11:18:27","extension":"xml","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":975,"visible":true,"origin":"","legend":"","description":"","filename":"JTRMD2523156Import.xml","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/07659b93a81513985cf06ca8.xml"},{"id":100676582,"identity":"b2a11d94-d891-4d20-9c24-03ab5883c8c2","added_by":"auto","created_at":"2026-01-20 11:20:56","extension":"xml","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":149939,"visible":true,"origin":"","legend":"","description":"","filename":"JTRMD25231560enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/9b766bdbfa98a41b4881bb5a.xml"},{"id":100676588,"identity":"3e5736e9-aa70-466c-993d-2de8b95686c1","added_by":"auto","created_at":"2026-01-20 11:21:12","extension":"jpg","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":388362,"visible":true,"origin":"","legend":"","description":"","filename":"Fig1MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/601f2ba2e22da28ad7634fbb.jpg"},{"id":100796502,"identity":"d944e5ef-0fd9-43b6-887a-27d0c7ad455b","added_by":"auto","created_at":"2026-01-21 13:43:44","extension":"jpg","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":354797,"visible":true,"origin":"","legend":"","description":"","filename":"Fig2MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/150273ed993db7f45121a741.jpg"},{"id":100676570,"identity":"87f9a6d6-e7e3-412f-8e38-0c4f93a424c4","added_by":"auto","created_at":"2026-01-20 11:20:22","extension":"jpg","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":359475,"visible":true,"origin":"","legend":"","description":"","filename":"Fig3MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/a335bb0f22350f1ca398ce64.jpg"},{"id":100676614,"identity":"1979f5c3-8724-4137-8c5c-0ea3a777b5f6","added_by":"auto","created_at":"2026-01-20 11:21:38","extension":"jpg","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":369907,"visible":true,"origin":"","legend":"","description":"","filename":"Fig4MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/ea4b3a08e575e5acfbbd477f.jpg"},{"id":100676313,"identity":"14c52cd4-8400-4444-b90f-8a4686f1dda0","added_by":"auto","created_at":"2026-01-20 11:18:27","extension":"jpg","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":534737,"visible":true,"origin":"","legend":"","description":"","filename":"Fig5MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/6c7af1f0a8e1a887a2fe779c.jpg"},{"id":100676174,"identity":"f692f8d1-ddeb-4a50-a1af-a0ebb3411b09","added_by":"auto","created_at":"2026-01-20 11:16:45","extension":"jpg","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":315672,"visible":true,"origin":"","legend":"","description":"","filename":"Fig6MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/de2527d6287631af57200a1a.jpg"},{"id":100676394,"identity":"9e7d840a-8f4a-4921-8e87-7ae0bfa54a1d","added_by":"auto","created_at":"2026-01-20 11:19:29","extension":"jpg","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":464144,"visible":true,"origin":"","legend":"","description":"","filename":"Fig7MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/ae78c2298c55055b31758e44.jpg"},{"id":100676586,"identity":"f0d19791-6b4a-477b-9fa6-91782aa43825","added_by":"auto","created_at":"2026-01-20 11:21:10","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5222176,"visible":true,"origin":"","legend":"","description":"","filename":"GAxgboost.png","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/0bd88e28f2be2af6cab1f8dc.png"},{"id":100676536,"identity":"b6b039d0-57e9-46df-abcb-becd08b0f13a","added_by":"auto","created_at":"2026-01-20 11:19:55","extension":"png","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":33815,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig1MLJTMdef.png","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/1e067f79fdf81c62490f34ac.png"},{"id":100676352,"identity":"ceeb5fe0-1a16-456a-a98c-911fe570c08a","added_by":"auto","created_at":"2026-01-20 11:19:10","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":43115,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig2MLJTMdef.png","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/9531d61ae33ee88fe3c63c10.png"},{"id":100676346,"identity":"f7468c69-eb3f-4029-9fc6-07d30c7e7e2c","added_by":"auto","created_at":"2026-01-20 11:19:01","extension":"png","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":71536,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig3MLJTMdef.png","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/4948fe83e9cf6b73a378e5cb.png"},{"id":100676567,"identity":"423eb40f-c747-4057-ac92-0d60cadaee81","added_by":"auto","created_at":"2026-01-20 11:20:18","extension":"png","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47851,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig4MLJTMdef.png","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/03dfde6706789e0b9ebf1526.png"},{"id":100676348,"identity":"fc14aa5a-f377-468f-92b8-53fc3a6c32b8","added_by":"auto","created_at":"2026-01-20 11:19:06","extension":"png","order_by":29,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":43791,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig5MLJTMdef.png","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/c2bbc109215e6c335baa7eff.png"},{"id":100676231,"identity":"9c8ebe28-603d-4a0e-a333-141674e37401","added_by":"auto","created_at":"2026-01-20 11:17:18","extension":"png","order_by":30,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":45029,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig6MLJTMdef.png","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/d12c8b09602281200ba06420.png"},{"id":100676580,"identity":"72dd0345-4d87-4227-adf2-7762ceea054e","added_by":"auto","created_at":"2026-01-20 11:20:53","extension":"png","order_by":31,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":61197,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig7MLJTMdef.png","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/1475a395c948f9241d5282fc.png"},{"id":100676196,"identity":"d38670f8-672e-4fd4-b6f9-d03f0470215a","added_by":"auto","created_at":"2026-01-20 11:16:57","extension":"png","order_by":32,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":291971,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineGAxgboost.png","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/6eb2771433350f9f50afd1ec.png"},{"id":100676581,"identity":"a31c07cf-33ce-4228-b621-d94622cd9f08","added_by":"auto","created_at":"2026-01-20 11:20:54","extension":"xml","order_by":33,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":143491,"visible":true,"origin":"","legend":"","description":"","filename":"JTRMD25231560structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/5dbd9e7f5bbc2b497ac19653.xml"},{"id":100676345,"identity":"05f7e0cd-b5a5-4f98-ab2a-fd9317714ca8","added_by":"auto","created_at":"2026-01-20 11:19:00","extension":"html","order_by":34,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":166079,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/8ae4e45996b0fc95a3fc8779.html"},{"id":100676281,"identity":"3e6e1bb1-9cdc-486c-9a5e-c82aafb5a92d","added_by":"auto","created_at":"2026-01-20 11:17:55","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":388362,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eForest Plot of Multivariable Predictors for ICI Response.\u003c/strong\u003e Forest plot of adjusted odds ratios and 95% confidence intervals for clinical, pathological, and molecular features in the logistic regression model for the immunotherapy response. Only the presence of lymph node metastasis reached statistical significance (OR \u0026gt; 2, P = 0.008), whereas all other tested variables (brain, liver, bone, lung, treatment type, sex, PD-L1 expression, and cfDNA) showed no significant individual effects.\u003c/p\u003e","description":"","filename":"Fig1MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/152e6c4b0fdf1519cfa602a0.jpg"},{"id":100676592,"identity":"c6fc9a72-950f-47e5-9ffd-ddef77f6fc00","added_by":"auto","created_at":"2026-01-20 11:21:14","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":354797,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eWorkflow for Germline Variant Processing and Predictive Modelling.\u003c/strong\u003e \u003cstrong\u003ea)\u003c/strong\u003e Schematic of the stepwise processing pipeline. Raw Whole-Exome Sequencing data were filtered for quality, aligned, and annotated to yield high-confidence germline variants. Feature selection involved removing highly correlated variants and prioritising those with significant clinical associations. \u003cstrong\u003eb)\u003c/strong\u003e Average ROC curves of the four analytical scenarios were modelled using the XGBoost and LASSO models.\u003c/p\u003e","description":"","filename":"Fig2MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/abb77c2666f7d7271aeb472f.jpg"},{"id":100676276,"identity":"497398de-0cde-46e4-b521-3d3d027c9624","added_by":"auto","created_at":"2026-01-20 11:17:50","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":359475,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eImpact of Germline Variants on Survival Outcomes.\u003c/strong\u003e Dumbbell Plot comparing Median Overall Survival (OS, left panel) and Progression-Free Survival (PFS, right panel) between variant carriers (red) and non-carriers (blue). The analysis included the top germline variants prioritised by the machine learning model. P-values derived from the log-rank test are displayed for significant associations. Consistent with the multivariate analysis, variants in \u003cem\u003eCCZ1B\u003c/em\u003e, and \u003cem\u003ePHLDB1\u003c/em\u003e, were associated with extended survival, whereas \u003cem\u003ePDE4E\u003c/em\u003e, \u003cem\u003eSLC6A16\u003c/em\u003e, \u003cem\u003eOR10H5 \u003c/em\u003eand \u003cem\u003eSIGLEC11\u003c/em\u003e were linked to significantly reduced survival times.\u003c/p\u003e","description":"","filename":"Fig3MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/e19c311090a2a6a20d8cd18c.jpg"},{"id":100676597,"identity":"010d01ff-05b2-40a5-9b92-0608f0d0501b","added_by":"auto","created_at":"2026-01-20 11:21:22","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":369907,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eImpact of Germline Variants on Survival Outcomes.\u003c/strong\u003e Dumbbell Plot comparing Median Overall Survival (OS, left panel) and Progression-Free Survival (PFS, right panel) between variant carriers (red) and non-carriers (blue). The analysis included the top germline variants prioritised by the machine learning model. P-values derived from the log-rank test are displayed for significant associations. Consistent with the multivariate analysis, variants in \u003cem\u003eCCZ1B\u003c/em\u003e, and \u003cem\u003ePHLDB1\u003c/em\u003e, were associated with extended survival, whereas \u003cem\u003ePDE4E\u003c/em\u003e, \u003cem\u003eSLC6A16\u003c/em\u003e, \u003cem\u003eOR10H5 \u003c/em\u003eand \u003cem\u003eSIGLEC11\u003c/em\u003e were linked to significantly reduced survival times.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"Fig4MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/a7fdad18c6d9697e8dae9050.jpg"},{"id":100676398,"identity":"7ebd073d-a083-4a5a-bded-fe2a49eff461","added_by":"auto","created_at":"2026-01-20 11:19:48","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":534737,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMultivariable Logistic Regression of Clinical and Germline Predictors of Immunotherapy Resistance.\u003c/strong\u003e Forest plots displaying the independent association of clinical factors (metastatic sites, PD-L1 expression, ECOG status, treatment type) and specific germline variants with poor clinical response. Separate multivariable models were generated for each prioritised variant: \u003cem\u003eSLC6A16\u003c/em\u003e, \u003cem\u003ePDE4E\u003c/em\u003e, \u003cem\u003eARHGAP22\u003c/em\u003e, and \u003cem\u003eOR10H5\u003c/em\u003e. Red squares indicate statistically significant associations (P \u0026lt; 0.05). In all models, lymph node metastasis and the presence of the indicated germline variants were consistently identified as independent predictors of treatment resistance (OR \u0026gt; 1).\u003c/p\u003e","description":"","filename":"Fig5MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/5ce1692a9590ece8aaacb1ea.jpg"},{"id":100676587,"identity":"6076e163-a00e-4eee-9750-ec0d3e00f411","added_by":"auto","created_at":"2026-01-20 11:21:12","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":315672,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eClinical–Germline Risk Score Stratifies Progression-Free Survival.\u003c/strong\u003e Kaplan–Meier survival curves for \u003cstrong\u003e(a)\u003c/strong\u003ethe total patient cohort and \u003cstrong\u003e(b)\u003c/strong\u003e the lung adenocarcinoma (LUAD) subgroup. Patients were stratified into “low” (orange) and “high” (green) risk categories using a composite score that integrated the prioritised germline variants and lymph node metastasis status. Risk tables display the number of patients at risk at each time point in the study. The composite score significantly stratified outcomes in both the general and histological subgroups (log-rank P \u0026lt; 0.001 and P = 0.012, respectively).\u003c/p\u003e","description":"","filename":"Fig6MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/6ad230dab0b18dd8413424dd.jpg"},{"id":100676585,"identity":"b1da51b0-f361-4557-a2ef-dff0a6e2ca5d","added_by":"auto","created_at":"2026-01-20 11:21:08","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":464144,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eInternal and External Validation of the Risk Score.\u003c/strong\u003e \u003cstrong\u003e(a-b)\u003c/strong\u003e Internal Temporal Validation: PFS stratification in the temporal hold-out validation set for \u003cstrong\u003e(a)\u003c/strong\u003e all subtypes and \u003cstrong\u003e(b)\u003c/strong\u003e LUAD patients, applying the original model cutoffs. \u003cstrong\u003e(c-d)\u003c/strong\u003eExternal Exploratory Validation: Kaplan-Meier curves for \u003cstrong\u003e(c)\u003c/strong\u003e an independent NSCLC cohort (cBioPortal, somatic data proxy) and \u003cstrong\u003e(d)\u003c/strong\u003e an independent melanoma cohort treated with ICIs. Although statistical significance was not reached in the external somatic/melanoma cohorts, the separation of curves followed the trend observed in the discovery cohort.\u003c/p\u003e","description":"","filename":"Fig7MLJTMdef.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/cf7d3f0495d70b882d884320.jpg"},{"id":100798031,"identity":"2790b58b-df97-40e1-9e23-1c71442a0bb3","added_by":"auto","created_at":"2026-01-21 13:52:17","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4416281,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/c87de08c-d741-4bb7-858a-970dd8fff72c.pdf"},{"id":100676307,"identity":"3b173d17-fc16-4a09-8478-4b1737176bc5","added_by":"auto","created_at":"2026-01-20 11:18:25","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"supplement","size":5222176,"visible":true,"origin":"","legend":"","description":"","filename":"GAxgboost.png","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/a7e153420096990e7cbe0482.png"},{"id":100676578,"identity":"a37b5ad8-17dc-40bb-973d-51962fd83b5f","added_by":"auto","created_at":"2026-01-20 11:20:49","extension":"docx","order_by":14,"title":"","display":"","copyAsset":false,"role":"supplement","size":19216,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterialtable1.docx","url":"https://assets-eu.researchsquare.com/files/rs-8455214/v1/64d65e4053d70178d3d01fbb.docx"}],"financialInterests":"","formattedTitle":"Prospective Germline Exome and Machine Learning-Based Risk Score Identify Predictive and PrognosticBiomarkers of Immunotherapy Outcomes in Advanced Non-Small Cell Lung Cancer","fulltext":[{"header":"Background","content":"\u003cp\u003eNon-small cell lung cancer (NSCLC) represents approximately 85% of all lung cancer cases and continues to be the leading cause of cancer-related mortality worldwide [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Immune checkpoint inhibitors (ICIs), notably those targeting the PD-1 and PD-L1 pathways, have transformed the treatment paradigms for advanced NSCLC and significantly extended survival in select patients [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. However, only a subset of individuals achieves durable clinical responses, emphasising the need for robust biomarkers to guide optimal patient selection.\u003c/p\u003e \u003cp\u003eCurrently used tumour-based predictors, such as PD-L1 expression, tumour mutational burden, and microsatellite instability, show limited and inconsistent values across real-world populations [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Consequently, there is an urgent need to expand the biomarker repertoire beyond the tumour microenvironment. Recently, host-related factors, specifically germline genetic variations, have gained recognition as critical modulators of anti-tumour immune responses, influencing both treatment efficacy and immune-related toxicities [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Several retrospective and multi-omics studies have linked germline variants to immune phenotypes and clinical endpoints in ICI-treated NSCLC cohorts [\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. However, germline-informed predictive models have seldom been prospectively validated or integrated with clinical and tumour-derived data within real-world settings, limiting their clinical translation.\u003c/p\u003e \u003cp\u003eConcurrently, advances in machine learning (ML) and artificial intelligence (AI) have enabled the analysis of complex multidimensional clinicogenomic datasets, uncovering subtle and non-linear predictors of therapeutic efficacy that escape traditional statistical methods[\u003cspan additionalcitationids=\"CR12 CR13\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Despite this technological progress, the translation of germline genomics into clinically actionable decision-making tools remains unfulfilled.\u003c/p\u003e \u003cp\u003eBuilding on current evidence, we hypothesised that specific inherited variants are significantly associated with differential responses to ICIs in patients with advanced NSCLC and that machine learning-based integration of these signals facilitates robust risk stratification.\u003c/p\u003e \u003cp\u003eHere, we present a prospective, multicentre study that systematically integrates whole-exome germline sequencing, comprehensive clinical data, and state-of-the-art ML algorithms to develop and validate a composite risk score for immunotherapy benefit in NSCLC, aiming to advance biomarker-driven precision medicine.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy Design and Patient Cohort\u003c/h2\u003e \u003cp\u003e This prospective, multicentre study enrolled 117 adult patients (\u0026ge;\u0026thinsp;18 years) with histologically confirmed stage IIIB/IV NSCLC treated between January 2018 and June 2024 at two hospitals in M\u0026aacute;laga (Hospital Universitario Virgen de la Victoria and Hospital Regional Universitario de M\u0026aacute;laga), Spain. Eligible patients were scheduled to receive immune checkpoint inhibitor (ICI)-based therapy, either as monotherapy or in combination with chemotherapy. Key exclusion criteria included prior ICI exposure and a history of other malignancies (within five years, except for non-melanoma skin cancer). All patients provided written informed consent. The study was approved by the regional ethics committee and conducted in accordance with the Declaration of Helsinki. To ensure patient privacy during machine learning analysis, all clinical and genomic data were pseudonymized and stored on secure servers, in strict compliance with the General Data Protection Regulation (GDPR) (EU) 2016/679\u003c/p\u003e \u003cp\u003eSample Collection and Clinical Data\u003c/p\u003e \u003cp\u003ePeripheral blood samples were collected in EDTA tubes before the first ICI administration. Germline DNA was extracted using the QIAamp DNA Mini Kit (Qiagen) and quantified with Qubit dsDNA BR Assay Kit (Thermo Fisher Scientific). Comprehensive clinical data were collected from electronic medical records, including age, sex, histological subtype (adenocarcinoma [LUAD], squamous cell carcinoma [LUSC], or others), TNM stage (IIIB vs. IV), treatment regimen, PD-L1 expression status, metastatic sites, and ECOG performance status. Treatment efficacy was evaluated according to the RECIST v1.1 criteria [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e], with tumour assessment performed every 8\u0026ndash;12 weeks according to clinical practice. Clinical outcomes were categorised as Complete Response (CR), Partial Response (PR), Stable Disease (SD), or Progressive Disease (PD). [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eWhole-Exome Library Preparation and Sequencing\u003c/h3\u003e\n\u003cp\u003eLibraries were prepared from 50 ng of germline DNA using the Twist Human Core Exome 2.0 Kit (Twist Bioscience), which targets a total of 33 Mb of coding regions. The quality of the libraries was assessed using an Agilent 2100 Bioanalyzer. Sequencing was performed on an Illumina NovaSeq 6000 platform (2 \u0026times; 150 bp), achieving a mean coverage of \u0026gt;\u0026thinsp;30x. FastQC v0.11.9 [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] and Fastp v0.39. [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] were used for raw data quality control and adapter trimming, respectively.\u003c/p\u003e\n\u003ch3\u003eBioinformatics Processing and Variant Annotation\u003c/h3\u003e\n\u003cp\u003eSequencing reads were aligned to the GRCh38 human reference genome using the BWA-MEM. Duplicates were marked, and base quality scores were recalibrated using Picard and GATK. Germline variants were called with GATK HaplotypeCaller, filtered per GATK best practices [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] (minimum depth\u0026thinsp;\u0026gt;\u0026thinsp;30x, QD\u0026thinsp;\u0026gt;\u0026thinsp;2.0, FS\u0026thinsp;\u0026lt;\u0026thinsp;60.0, MQ\u0026thinsp;\u0026gt;\u0026thinsp;40.0), and restricted to the 40\u0026ndash;60% allelic fraction to minimise clonal haematopoietic artefacts. Functional annotation was performed using CRAVAT [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] and gnomAD v3.1.2 [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] for allele frequency analysis.\u003c/p\u003e \u003cp\u003eVariants were classified according to pathogenicity, following The American College of Medical Genetics and Genomics (ACMG) recommendations [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Only rare (MAF\u0026thinsp;\u0026lt;\u0026thinsp;1%), exonic, splice-site, and high-impact intronic variants were selected. To reduce data redundancy and minimise multicollinearity, we applied a pairwise correlation filter; for variant pairs with high correlation (Pearson\u0026rsquo;s r\u0026thinsp;\u0026ge;\u0026thinsp;0.70), only one representative variant was retained for downstream analysis. Variant annotation was blinded to the clinical outcomes.\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eAssociations between genetic variants and clinical benefits were assessed using Fisher\u0026rsquo;s exact test and logistic regression with FDR correction. To align with the concept of disease control, the outcome variable was dichotomised into responders (patients achieving CR, PR, or SD) and non-responders (patients with PD). Progression-free survival (PFS) and overall survival (OS) were compared using Kaplan\u0026ndash;Meier curves and log-rank tests. Analyses were run in R v4.2.0 (packages: tidyverse, survival, xgboost). Statistical significance was set at P\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eMachine Learning and Predictive Modelling\u003c/h3\u003e\n\u003cp\u003eTo ensure robust phenotypic separation, we first defined a cohort of 'extreme responders' (non-responders\u0026thinsp;\u0026lt;\u0026thinsp;3 months vs. responders\u0026thinsp;\u0026gt;\u0026thinsp;12 months). Using this cohort, we considered four modelling scenarios: (1) all selected germline variants, (2) exonic-only variants, (3) all variants per LUAD subgroup, and (4) exonic variants of LUAD only.\u003c/p\u003e \u003cp\u003eFor model construction, the dataset corresponding to these extreme phenotypes was split into training (70%) and test (30%) sets, and 1,000 iterations were performed to evaluate model stability and feature importance. Within each training set, hyperparameter tuning used stratified 5-fold cross-validation to minimise data leakage.\u003c/p\u003e \u003cp\u003eTo minimise overfitting and ensure rigorous evaluation, model construction was performed on a training set comprising 70% of the extreme responder population. For internal validation, we employed a temporal hold-out strategy, reserving the subset of extreme responder patients recruited during the final 12 months of the study period (n\u0026thinsp;=\u0026thinsp;37). This temporal split mimics a prospective clinical scenario in which the model is tested on future patients, providing a more robust estimate of predictive performance than random splitting.\u003c/p\u003e \u003cp\u003eThe optimised model from each iteration was evaluated on the test set to assess predictive performance (AUC). The AUC values reflect the predictive accuracy of the immunotherapy response within the held-out test sets to ensure generalisability. Feature selection, model stability, and variable importance metrics were collated across the iterations. Subsequent survival modelling grouped patients according to genotype status (variant carriers vs. non-carriers) and compared outcomes using the log-rank test. To define the optimal thresholds, the surv_cutpoint function from the \u003cem\u003eSurvminer\u003c/em\u003e R package was used to identify the cutoff value that maximised the log-rank statistic. Based on the biological premise that mechanisms facilitating initial immune recognition (response) are essential for durable disease control, we hypothesised that robust response predictors would also stratify survival outcomes. Consequently, two composite risk scores were generated by integrating variables that demonstrated consistency across both the ML response model and univariate PFS analyses: (i) The Global Clinical-Germline Score (all patients), comprising six adverse predictors (five molecular, one clinical) and three favourable predictors (two molecular, one clinical), and (ii) The LUAD-Specific Score (adenocarcinoma only), comprising three adverse predictors and three favourable predictors. Adverse predictors (OR\u0026thinsp;\u0026gt;\u0026thinsp;1) were incorporated as penalties by adding their log(OR) values, whereas favourable predictors (OR\u0026thinsp;\u0026lt;\u0026thinsp;1) were treated as protective factors by subtracting log(OR) values. This weighting strategy reflects the magnitude of the association between each variable and the clinical outcome. The thresholds for the high- and low-risk groups were determined using an optimal cutoff based on (a) a minimum subgroup size of 10% and (b) maximisation of survival curve separation. To assess threshold robustness, a nonparametric bootstrap procedure (1,000 resamples) was used to estimate the 95% confidence interval of the optimal cutoff distribution.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eExternal Validation Cohorts\u003c/h2\u003e \u003cp\u003eTo evaluate the robustness and generalisability of the derived clinical-germline risk score, we analysed two independent validation cohorts. First, to assess the performance of the gene signature in an external NSCLC population and given the scarcity of public datasets with matched germline-clinical immunotherapy data, we queried cBioPortal for Cancer Genomics. We selected the dataset from Rizvi et al. 2015/Hellmann et al. 2018, comprising 75 patients treated with immune checkpoint inhibitors [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. As this dataset primarily provides somatic sequencing data, validation was restricted to assessing the prognostic value of the identified gene signature at the gene level. Second, to explore the utility of the score across different tumour types (cross-tumour validation), we included an independent in-house cohort of 19 patients with metastatic melanoma. These patients were treated with anti-PD-1/PD-L1 monotherapy at Hospital Regional de Malaga, following the same inclusion criteria and ethical standards as those of the main cohort. Germline sequencing and clinical data collection for this subset followed the pipeline described above.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eData and Code Availability\u003c/h3\u003e\n\u003cp\u003eRaw sequencing data, genotype matrices, and scripts can be obtained from the corresponding authors upon reasonable request and institutional approval.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003ePatients\u0026rsquo; characteristics\u003c/h2\u003e \u003cp\u003eA total of 117 patients with advanced NSCLC were enrolled. The median age was 67.5 years (range: 46\u0026ndash;85 years), and 74.4% were male. Histologically, lung adenocarcinoma (LUAD) accounted for 59.8% of the cases, while 30.8% were squamous cell carcinomas (LUSC). Although 19.1% of patients were initially diagnosed with locally advanced disease (Stage IIIB), all presented with metastatic progression or Stage IV disease at the time of ICI initiation. Among those with metastatic spread, the most frequent sites were the lungs (51.3%) and the lymph nodes (47.9%). ICI monotherapy was administered in 25.6% of cases, with combination chemo-ICI or other regimens administered in 74.4% of cases. At the first evaluation per RECIST v1.1, the Objective Response Rate (CR or PR) was 17.1% in evaluable patients, whereas 65% achieved Disease Control (CR, PR, or SD) by 3 months. The detailed cohort characteristics are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003eDemographic and Clinical Features of the Study Population\u003c/b\u003e Summary of the baseline characteristics of the 117 patients with advanced NSCLC included in the study.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCategory\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003en (117)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e%\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSex\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFemale (F)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e25.6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMale (M)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e74.4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHistology\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLUAD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e59.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLUSC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e36\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e30.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNSCLC NS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOther\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3.4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eStage\u0026nbsp;at diagnosis\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIIIB (locally advanced)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e19.13\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eIV (Metastatic)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e72.17\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePD-L1 expression\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e53.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u0026ndash;49%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e29.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u0026gt;\u0026thinsp;50%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNot evaluated\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTherapy type\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eICI monotherapy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e25.6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCT\u0026thinsp;+\u0026thinsp;ICI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e43\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e36.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCT\u0026thinsp;+\u0026thinsp;ICI\u0026thinsp;+\u0026thinsp;Others\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e33.3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eImmunotherapy combo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRECIST 1st evaluation (8\u0026ndash;12 weeks)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eComplete Response (CR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4.3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePartial Response (PR)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e12.8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStable Disease (SD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.6\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eProgressive Disease (PD)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e9.4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNot evaluable\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e70.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eClinical Benefit At 3 months (Responder vs Non-Responders)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eResponders\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e65.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNon-responders\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e35.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eClinical Benefit At 12 months (Responder vs Non-Responders)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eResponders\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e33.3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNon-responders\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e59.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eM1 Metastasis\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBrain\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e12.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLiver\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e12\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e10,25\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e34,18\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLung\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e51,28\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSkin\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e25,64\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLymph node\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e47,86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003e* Abbreviations: CT, chemotherapy; ICI, immune checkpoint inhibitor; IT, immunotherapy; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma; NSCLC NOS, non-small cell lung cancer not otherwise specified; NE, not evaluable; M1, distant metastasis. \u0026uml;Other\" histology includes large cell carcinoma and adenosquamous carcinoma. M1 sites refer to distant metastases confirmed at the baseline.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eClinical Variables associate with Immunotherapy Response\u003c/h2\u003e \u003cp\u003eMultivariable logistic regression confirmed that lymph node metastasis was the sole independent clinical predictor of poor response to ICIs (OR 7.9, 95% CI 1.8\u0026ndash;43.3, P\u0026thinsp;=\u0026thinsp;0.008). Given that all patients were treated in the advanced/metastatic setting, the tumour stage was not included as a covariate. Regarding the metastatic burden, no specific organ involvement (brain, liver, bone, lung, or skin) other than lymph nodes was significantly associated with the outcome, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Neither treatment modality (ICI monotherapy vs. combination), sex, PD-L1 expression strata, nor ECOG performance status were significantly associated with clinical benefits.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eResponse Germline Variant Discovery and Model Performance\u003c/h2\u003e \u003cp\u003eTo identify germline predictors of immunotherapy response, we implemented a comprehensive variant filtering and prioritisation pipeline (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003ea). From an initial pool of 108,944 called variants, strict quality control and linkage disequilibrium pruning (Pearson\u0026rsquo;s r\u0026thinsp;\u0026lt;\u0026thinsp;0.70) yielded a refined set of candidates. Univariate statistical filtering further narrowed this pool to those showing the strongest association with clinical benefit.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe predictive performance of the machine learning models was evaluated across four analytical scenarios. The XGBoost model utilising all exonic variants achieved the highest accuracy, with a cross-validated median area under the ROC curve (AUC) of 0.845 (IQR 0.742\u0026ndash;0.879) in the held-out test set (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eb). Comparative modelling consistently demonstrated the superiority of XGBoost over LASSO regression across all datasets. For the LUAD-specific cohort, XGBoost achieved an AUC of 0.82, compared to 0.56 for LASSO. Similarly, in models restricted to exonic variants within LUAD, XGBoost maintained a robust performance (AUC 0.77) compared to LASSO (AUC 0.72).\u003c/p\u003e \u003cp\u003eFeature importance analysis revealed distinct genetic signatures depending on the histological context. In the global analysis (all patients), the most recurrently selected loci (\u0026gt;\u0026thinsp;70% of iterations) included variants in \u003cem\u003eOR51M1, C22orf34, DENND2B, SLC6A16, ZNF148, SIGLEC11, ZNF454\u003c/em\u003e, and \u003cem\u003ePDE4E\u003c/em\u003e genes. Conversely, the analysis focused specifically on the LUAD subgroup highlighted distinct drivers, identifying \u003cem\u003eARAP3, CCZ1B, ARHGAP22\u003c/em\u003e, and \u003cem\u003eOR10H5\u003c/em\u003e as significant contributors. The frequent selection of these variants suggests their potential role in germline-mediated modulation of the immune response.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Validation and Prognostic Characterization of Prioritised Variants\u003c/h2\u003e \u003cp\u003eTo individually validate the germline features prioritised by the best-performing XGBoost model (exonic variants) and quantify their effect sizes, we performed univariate logistic regression analysis. This post-hoc analysis confirmed that several germline variants were significantly associated (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05) with clinical benefits (responders vs. non-responders). Specifically, variants chr19:49309146 G\u0026thinsp;\u0026gt;\u0026thinsp;T (\u003cem\u003eSLC6A16\u003c/em\u003e), chr19:49960436_\u0026gt;T (\u003cem\u003eSIGLEC11\u003c/em\u003e), chr5:178964901 A\u0026thinsp;\u0026gt;\u0026thinsp;C (\u003cem\u003eZNF454\u003c/em\u003e), chr5:60460204 A\u0026thinsp;\u0026gt;\u0026thinsp;G (\u003cem\u003ePDE4E\u003c/em\u003e), chr10:48451594 T\u0026thinsp;\u0026gt;\u0026thinsp;C (\u003cem\u003eARHGAP22\u003c/em\u003e), and chr19:15794719 C\u0026thinsp;\u0026gt;\u0026thinsp;A (\u003cem\u003eOR10H5\u003c/em\u003e) were associated with a higher likelihood of non-response (resistance). In contrast, chr11:8766678 A\u0026thinsp;\u0026gt;\u0026thinsp;C (\u003cem\u003eDENND2B\u003c/em\u003e), chr11:118643907 C\u0026thinsp;\u0026gt;\u0026thinsp;T (\u003cem\u003ePHLDB1\u003c/em\u003e), chr5:141671669 C\u0026thinsp;\u0026gt;\u0026thinsp;T (\u003cem\u003eARAP3\u003c/em\u003e), and chr7:6822249 G\u0026thinsp;\u0026gt;\u0026thinsp;C (\u003cem\u003eCCZ1B\u003c/em\u003e) were associated with a positive treatment outcome (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e\u003cb\u003ePrioritised Germline Variants and Statistical Associations with Immunotherapy Outcomes\u003c/b\u003e List of the top germline variants selected by machine learning models and univariate logistic regression, showing odds ratios, confidence intervals, and p-values for immunotherapy response, progression-free survival (PFS), and overall survival (OS). Genes, variant positions, variant types (synonymous, non-synonymous, and intronic), and relevant molecular functions were annotated for each locus\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c6\" namest=\"c4\"\u003e \u003cp\u003eUnivariante\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePFS\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eOS\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGene\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eVariant (chr:position/ref/alt)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eVariant Type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eOR\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e95% CI\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ep-val\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eLog-Rank P-val\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eLog-Rank P-val\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eOR51M1\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr11:5390165:C\u0026thinsp;\u0026gt;\u0026thinsp;A\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNon-synonymous\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(0.7\u0026ndash;14.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.96\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDENND2B\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr11:8766678:A\u0026thinsp;\u0026gt;\u0026thinsp;C\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIntronic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(0.08 -0,86)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.03*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.067\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.22\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSLC6A16\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr19:49309146:G\u0026thinsp;\u0026gt;\u0026thinsp;T\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIntronic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(1,73- 15.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.004**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.0048**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.0057**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSIGLEC11\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr19:49960436:_:T\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIntronic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e8.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(2.5\u0026ndash;41)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.002**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.03*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.00021**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eC22orf34\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr22:49624397:G\u0026thinsp;\u0026gt;\u0026thinsp;A\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIntronic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(0.79- 26.)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.043*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eZNF148\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr3:125232977:C\u0026thinsp;\u0026gt;\u0026thinsp;T\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNon-synonymous\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(0,58\u0026thinsp;\u0026minus;\u0026thinsp;8.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eZNF454\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr5:178964901:A\u0026thinsp;\u0026gt;\u0026thinsp;C\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNon-synonymous\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(1.28\u0026ndash;22.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.03*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.17\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePDE4E\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr5:60460204:A\u0026thinsp;\u0026gt;\u0026thinsp;G\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIntronic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6.13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(2.5\u0026ndash;21.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.002**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.0051**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.0001**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePHLDB1\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr11:118643907:C\u0026thinsp;\u0026gt;\u0026thinsp;T\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNon-synonymous\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(0.04-0,44)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.001**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.036*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.011*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eARAP3\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr5:141671669:C\u0026thinsp;\u0026gt;\u0026thinsp;T\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNon-synonymous\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.22\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(0,06- -0,63)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.007**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.085\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.0001**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eARHGAP22\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr10:48451594:T\u0026thinsp;\u0026gt;\u0026thinsp;C\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIntronic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(1.32\u0026ndash;10.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.01**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.049*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.05*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCCZ1B\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr7:6822249:G\u0026thinsp;\u0026gt;\u0026thinsp;C\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eIntronic\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(0.04\u0026ndash;0.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.001**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.0016**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.00031**\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eOR10H5\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003echr19:15794719:C\u0026thinsp;\u0026gt;\u0026thinsp;A\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNon-synonymous\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e(2.03\u0026ndash;33.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.005**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.0075**\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.028*\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eFurthermore, we assessed progression-free survival (PFS) to corroborate these findings. The median PFS was compared between patients harbouring the variant (variant-positive) and those with the reference genotype (wild type). Consistent with the logistic regression results, the \u003cem\u003eCCZ1B\u003c/em\u003e variant (chr7:6822249 G\u0026thinsp;\u0026gt;\u0026thinsp;C) was associated with significantly extended PFS (21 vs. 9 months; P\u0026thinsp;=\u0026thinsp;0.0016), aligning with its protective odds ratio OR (0.14). A similar survival benefit was observed for \u003cem\u003ePHLDB1\u003c/em\u003e (chr11:118643907 C\u0026thinsp;\u0026gt;\u0026thinsp;T; 12 vs. 8 months; P\u0026thinsp;=\u0026thinsp;0.036).\u003c/p\u003e \u003cp\u003eConversely, the presence of variants in \u003cem\u003eSLC6A16\u003c/em\u003e (chr19:49309146 G\u0026thinsp;\u0026gt;\u0026thinsp;T ) (P\u0026thinsp;=\u0026thinsp;0.0048), \u003cem\u003eOR10H5(\u003c/em\u003echr19:15794719 C\u0026thinsp;\u0026gt;\u0026thinsp;A) (P\u0026thinsp;=\u0026thinsp;0.0075), \u003cem\u003ePDE4E\u003c/em\u003e (chr5:60460204 A\u0026thinsp;\u0026gt;\u0026thinsp;G ) (P\u0026thinsp;=\u0026thinsp;0.0051), \u003cem\u003eARHGAP22\u003c/em\u003e (chr10:48451594 T\u0026thinsp;\u0026gt;\u0026thinsp;C) (P\u0026thinsp;=\u0026thinsp;0.049), and \u003cem\u003eSIGLEC11\u003c/em\u003e (chr19:49960436_\u0026gt;T )(P\u0026thinsp;=\u0026thinsp;0.03) was associated with significantly reduced PFS compared to non-carriers. Variants in \u003cem\u003eDENND2B\u003c/em\u003e (chr11:8766678 A\u0026thinsp;\u0026gt;\u0026thinsp;C\u003cem\u003e)\u003c/em\u003e and \u003cem\u003eARAP3\u003c/em\u003e (chr5:141671669 C\u0026thinsp;\u0026gt;\u0026thinsp;T) showed a trend toward PFS benefits (P\u0026thinsp;=\u0026thinsp;0.067 and P\u0026thinsp;=\u0026thinsp;0.085, respectively), although the differences were not significant.\u003c/p\u003e \u003cp\u003eFinally, we evaluated the Overall Survival (OS). Most notably, the \u003cem\u003eCCZ1B\u003c/em\u003e (chr7:6822249 G\u0026thinsp;\u0026gt;\u0026thinsp;C\u003cem\u003e)\u003c/em\u003e variant conferred a substantial survival advantage (median OS, 22 vs. 7 months; P\u0026thinsp;\u0026lt;\u0026thinsp;0.01). While \u003cem\u003ePHLDB1\u003c/em\u003e chr11:118643907 C\u0026thinsp;\u0026gt;\u0026thinsp;T and \u003cem\u003eARAP3\u003c/em\u003e (chr5:141671669 C\u0026thinsp;\u0026gt;\u0026thinsp;T) variants achieved significance for OS (despite missing the PFS threshold), variants in \u003cem\u003ePDE4E\u003c/em\u003e (chr5:60460204 A\u0026thinsp;\u0026gt;\u0026thinsp;G), \u003cem\u003eSLC6A16\u003c/em\u003e (chr19:49309146 G\u0026thinsp;\u0026gt;\u0026thinsp;T), \u003cem\u003eSIGLEC11 (\u003c/em\u003echr19:49960436_\u0026gt;T), \u003cem\u003eARHGAP22 (\u003c/em\u003echr10:48451594 T\u0026thinsp;\u0026gt;\u0026thinsp;C), and \u003cem\u003eOR10H5\u003c/em\u003e (chr19:15794719 C\u0026thinsp;\u0026gt;\u0026thinsp;A ) were associated with a marked decline in OS. Interestingly, the \u003cem\u003eC22orf34\u003c/em\u003e variant (chr22:49624397 G\u0026thinsp;\u0026gt;\u0026thinsp;A), which was not significant in the response or PFS analyses, was identified as a novel predictor of poor OS (P\u0026thinsp;=\u0026thinsp;0.043). The representative Kaplan\u0026ndash;Meier curves are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eMultivariable Analysis of Prioritized Germline and Clinical Data\u003c/h2\u003e \u003cp\u003eTo determine the independent predictive value of the identified biomarkers, we conducted a comprehensive multivariable logistic regression analysis. This model integrated the germline variants selected by the XGBoost algorithm, specifically those showing consistency in the univariate response and progression-free survival (PFS) assessments, along with key clinical covariates (sex, histology, PD-L1 expression, treatment type, and metastatic sites). In all model iterations, lymph node metastasis consistently emerged as a robust independent predictor of resistance, yielding Odds Ratios (OR) for non-response ranging from approximately 5.9 to 21.0 (P\u0026thinsp;\u0026lt;\u0026thinsp;0.05) depending on the variant combination (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). Regarding germline features, six variants maintained a significant correlation with treatment outcomes after adjusting for clinical factors. Variants in \u003cem\u003eSLC6A16\u003c/em\u003e (chr19:49309146 G\u0026thinsp;\u0026gt;\u0026thinsp;T), \u003cem\u003ePDE4E\u003c/em\u003e (chr5:60460204 A\u0026thinsp;\u0026gt;\u0026thinsp;G), and \u003cem\u003eARHGAP22\u003c/em\u003e(chr10:48451594 T\u0026thinsp;\u0026gt;\u0026thinsp;C) were confirmed as independent predictors of poor response. Patients carrying these variants experienced a significantly higher likelihood of resistance and shorter median PFS.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eConversely, the \u003cem\u003ePHLDB1\u003c/em\u003e(chr11:118643907C\u0026thinsp;\u0026gt;\u0026thinsp;T), \u003cem\u003eCCZ1B\u003c/em\u003e (chr7:6822249G\u0026thinsp;\u0026gt;\u0026thinsp;C), and \u003cem\u003eOR10H5\u003c/em\u003e (chr19:15794719C\u0026thinsp;\u0026gt;\u0026thinsp;A) variants were independently associated with favourable outcomes, exhibiting protective odds ratios in the regression model, and correlated with extended median PFS in carriers (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eNo consistent independent associations were found for other clinical factors, including distant metastases, immunotherapy regimen, sex, PD-L1 expression, or ECOG status. Consequently, lymph node involvement and the identified germline profile were the most reliable independent indicators of the clinical benefit. Notably, these predictive effects were observed for both non-synonymous and intronic variants, suggesting diverse functional mechanisms of immune modulation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eDevelopment and Validation of the Clinical-Germline Prognostic Score\u003c/h2\u003e \u003cp\u003eFinally, to translate these findings into a clinical tool, we constructed a composite risk score by integrating the prioritised germline variants with lymph node status, weighting each variable by its corresponding regression coefficient (log-odds ratio). Patients were stratified into high- and low-risk groups based on the optimal cutoff determined by the surv_cutpoint algorithm.\u003c/p\u003e \u003cp\u003eKaplan\u0026ndash;Meier analysis demonstrated a robust distinction in clinical outcomes between the strata. In the overall cohort, individuals in the low-risk category experienced significantly longer progression-free survival (PFS) than those in the high-risk group (median 18.0 vs. 7.0 months; log-rank P\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003ea). This stratification power was maintained in the LUAD subgroup (median PFS 19.0 vs. 5.0 months; P\u0026thinsp;=\u0026thinsp;0.012) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eb).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eValidation in Independent Cohorts\u003c/h2\u003e \u003cp\u003eTo assess the robustness of the score, we first applied a temporal hold-out strategy and tested the model on a subset of patients recruited during the final study period (n\u0026thinsp;=\u0026thinsp;22). Using the predefined cutoff, the score successfully stratified PFS (P\u0026thinsp;=\u0026thinsp;0.002; median 18 vs. 4 months), confirming internal stability (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ea-b).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eSubsequently, we explored the score's utility in external settings using an independent NSCLC cohort (cBioPortal[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] ) and an in-house melanoma cohort treated with ICIs. Since only somatic sequencing data were available for the external NSCLC cohort, we evaluated the gene-level impact of the signature. Although survival differences did not reach statistical significance, likely due to biological distinctions between somatic and germline alterations, similar stratification trends were observed, with low-risk patients consistently showing superior survival curves. This suggests that the identified gene pathways may play a conserved role in modulating immunotherapy outcomes across contexts (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003ec-d).\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur study provides new evidence that germline variants, when analysed using advanced machine learning models, deliver valuable predictive information for immunotherapy response and outcomes in non-small cell lung cancer (NSCLC) [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Crucially, the robustness of these findings was supported by rigorous internal temporal validation and further explored in independent external cohorts, reinforcing the potential generalisability of the identified biomarkers. Unlike tumour-restricted or PD-L1-based biomarkers, inherited genomic variations appear to play an active role in determining therapeutic response, reflecting constitutional influences on host immune competence and tumour-immune interactions. Variants within genes such as \u003cem\u003eCCZ1B, SIGLEC11, PDE4E, ARHGAP22, PHLDB1, SLC6A16\u003c/em\u003e, and \u003cem\u003eOR10H5\u003c/em\u003e were consistently associated with progression-free survival (PFS), supporting the notion that the germline genome encodes mechanisms that modulate immune activation or resistance.\u003c/p\u003e \u003cp\u003eThrough an agnostic approach, the machine learning algorithm identified several genes associated with immunotherapy response, some of which are linked to tumourigenic processes or have recently been implicated in pathways potentially regulating immunotherapy response in cancer. For example, \u003cem\u003eSIGLEC11\u003c/em\u003e is a membrane receptor that negatively regulates the immune response by recognising and binding to sialoglycans on tumour cells, facilitating immune evasion and metastasis via the suppression of anti-tumour immunity [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Other relevant genes include \u003cem\u003ePDE4E\u003c/em\u003e, a cAMP-specific phosphodiesterase linked to immunotherapy efficacy in LUAD [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e], as well as \u003cem\u003eSLC6A16\u003c/em\u003e, whose elevated expression in tumour cells has been shown to promote immunotherapy resistance by impairing CD8⁺ T cell function and reducing immune infiltration [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Mechanistic and Prognostic Implications Beyond these specific gene functions, these findings collectively emphasise the utility of genome-wide, hypothesis-free approaches to uncover germline-mediated immunotherapy response mechanisms. However, other genes identified by this model have not yet been associated with tumour development or immune-related pathways. These findings may reflect statistical noise or the involvement of previously unrecognised germline mechanisms, warranting further functional investigations [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAmong the favourable prognostic markers, \u003cem\u003eCCZ1B\u003c/em\u003e showed the greatest survival benefit. The \u003cem\u003eCCZ1B\u003c/em\u003e gene is involved in vesicular trafficking [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]; however, its specific role in boosting the cancer immune response remains to be elucidated. In contrast, variants in \u003cem\u003eOR10H5\u003c/em\u003e were associated with resistance. Although \u003cem\u003eOR10H5\u003c/em\u003e is canonically involved in olfaction, the identification of variants within this highly polymorphic gene family warrants caution because of the potential for mapping artefacts. While our stringent quality filtering minimises technical noise, we cannot exclude the possibility that this variant functions as a surrogate marker in a linkage disequilibrium with adjacent immune-regulatory loci. Future functional studies are essential to distinguish between the direct immunomodulatory role of \u003cem\u003eOR10H5\u003c/em\u003e and the passenger effect. Notably, however, its expression has previously been implicated in cancer biology; for instance, \u003cem\u003eOR10H5\u003c/em\u003e is part of a 19-gene prognostic signature in colorectal cancer, which correlates with improved patient survival [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Our data suggest a broader, context-dependent role in oncogenesis that warrants further investigation in NSCLC\u003c/p\u003e \u003cp\u003eAdditionally, the \u003cem\u003eC22orf34\u003c/em\u003e gene (a long non-coding RNA) was significantly associated with overall survival (OS) but not with PFS, suggesting that its prognostic relevance may be limited to long-term outcomes. Notably, \u003cem\u003eC22orf34\u003c/em\u003e has been previously associated with an increased risk of drug-induced interstitial lung disease (DIILD) in cancer patients [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e], highlighting a potential link to treatment-related toxicity.\u003c/p\u003e \u003cp\u003eBeyond single-variant effects, the integrated germline\u0026ndash;clinical score efficiently stratified patients according to PFS. The persistence of lymph node metastasis as an independent adverse factor underscores the multifactorial nature of resistance. Methodologically, the superior performance of XGBoost (AUC 0.845) over classical penalised regression approaches like LASSO (AUC 0.56\u0026ndash;0.72) demonstrates the critical advantage of ensemble learning in capturing non-linear genotype\u0026ndash;phenotype relationships. While traditional linear models often fail to account for the complex epistatic interactions inherent in genomic data, our ML approach effectively modelled these dependencies. This performance compares favourably with reported predictive metrics for standard biomarkers, such as PD-L1 alone, in real-world settings, suggesting that germline features capture a component of \"missing heritability\" not addressed by current tumour-only markers. Notably, predictive models using exomic data yielded slightly superior AUC compared to the entire variant set. This highlights the importance of feature selection, even when exonic variants directly encode protein-altering changes. The reduction of variables to the most informative subset improved model accuracy by excluding redundant or noisy features and minimising overfitting, a finding consistent with recent studies applying hybrid dimension reduction techniques to cancer genomics [\u003cspan additionalcitationids=\"CR31\" citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eNonetheless, this study had limitations. The single-country design and moderate sample size may limit generalisability and affect the effect estimates for rare variants. Despite strong internal validation and exploratory external validation cohorts, strictly matched germline replication studies in multi-ethnic NSCLC populations are needed to confirm these findings. A further limitation of our external validation is the reliance on somatic sequencing data. Standard somatic variant calling pipelines often filter out germline signals as 'normal' background; therefore, our analysis likely captured somatic hits in the same genes rather than the original germline variants. However, the fact that somatic alterations in these loci also stratify risk supports the hypothesis that these genes are functional bottlenecks in tumour-immune interactions, regardless of the genomic origin (germline vs. somatic) of the alteration. Additionally, functional studies are required to causally connect these germline variations with immune-modulatory mechanisms.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eOur findings establish a foundation for a comprehensive approach to discover immunotherapy biomarkers and stratify risk by integrating clinical and germline genetic data. This study illustrates how advanced AI-driven modelling can harness the predictive capabilities of inherited genetic variations, offering a pathway toward more personalised and precise immunotherapy strategies for lung cancer.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNSCLC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNon-small cell lung cancer\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLUAD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLung adenocarcinoma\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLUSC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLung squamous cell carcinoma\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eICI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eImmune checkpoint inhibitor\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePD-1\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eProgrammed cell death protein 1\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePD-L1\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eProgrammed death-ligand 1\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eTMB\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTumor mutational burden\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePFS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eProgression-free survival\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eOS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eOverall survival\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eComplete response\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePartial response\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eSD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eStable disease\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ePD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eProgressive disease\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eOR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eOdds ratio\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eCI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eConfidence Interval\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eDNA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eDeoxyribonucleic Acid\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRNA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRibonucleic Acid\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNGS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNext-generation sequencing\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eWES\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eWhole-exome sequencing\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eML\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMachine learning\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eArtificial intelligence\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eEDTA\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eEthylenediaminetetraacetic acid\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eECOG\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eEastern Cooperative Oncology Group\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eRECIST\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eResponse Evaluation Criteria in Solid Tumours\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eQD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eQuality by depth\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eVAF\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eVariant allele frequency\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eACMG\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAmerican College of Medical Genetics and Genomics\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eFDR\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eFalse discovery rate\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eReceiver operating characteristic\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eArea under the curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003ecAMP\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCyclic adenosine monophosphate\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was conducted in accordance with the Declaration of Helsinki and approved by the Provincial Research Ethics Committee of Málaga (PEIBA), reference number 26-10-2017. Written informed consent was obtained from all participants included in the study.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors have provided consent for publication.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and/or analysed during the current study are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by Consejería de Conocimiento, Investigación y Universidad, Junta de Andalucía under grant number PAIDI P21-01002. Funding: JO and IB holds ‘Nicolas Monardes’ research contracts from the Andalusian Regional Ministry Health (JO: C1-0003-2023)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors’ contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eJO, AR, IB: conceptualization , AR , JLO, MRG, GPL: methodology and software, EPR, JCB , JZ, AC: clinical data curation, AGH, MGB, LCFO, BMG: experimental procedures, formal analysis, JO, AR: writing-original draft preparation, JO, AGR, AR: writing-review \u0026amp; editing, JO, ARD: supervision. All authors have read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe thank the Picasso Supercomputer, University of Málaga (UMA), for providing the high-performance computing resources essential for performing the genomic and machine learning analyses presented in this study.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71:209\u0026ndash;49. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3322/caac.21660\u003c/span\u003e\u003cspan address=\"10.3322/caac.21660\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReck M, Rodr\u0026iacute;guez-Abreu D, Robinson AG, Hui R, Csőszi T, F\u0026uuml;l\u0026ouml;p A, et al. Pembrolizumab versus Chemotherapy for PD-L1\u0026ndash;Positive Non\u0026ndash;Small-Cell Lung Cancer. New Engl J Med Mass Med Soc. 2016;375:1823\u0026ndash;33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1056/NEJMoa1606774\u003c/span\u003e\u003cspan address=\"10.1056/NEJMoa1606774\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGandhi L, Rodr\u0026iacute;guez-Abreu D, Gadgeel S, Esteban E, Felip E, De Angelis F, et al. Pembrolizumab plus Chemotherapy in Metastatic Non-Small-Cell Lung Cancer. N Engl J Med. 2018;378:2078\u0026ndash;92. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1056/NEJMoa1801005\u003c/span\u003e\u003cspan address=\"10.1056/NEJMoa1801005\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWallenta Law J, Bapat B, Sweetnam C, Mohammed H, McBratney A, Izano MA, et al. Real-World Impact of Comprehensive Genomic Profiling on Biomarker Detection, Receipt of Therapy, and Clinical Outcomes in Advanced Non-Small Cell Lung Cancer. JCO Precis Oncol. 2024;8:e2400075. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1200/PO.24.00075\u003c/span\u003e\u003cspan address=\"10.1200/PO.24.00075\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHellmann MD, Ciuleanu T-E, Pluzanski A, Lee JS, Otterson GA, Audigier-Valette C, et al. Nivolumab plus Ipilimumab in Lung Cancer with a High Tumor Mutational Burden. N Engl J Med. 2018;378:2093\u0026ndash;104. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1056/NEJMoa1801946\u003c/span\u003e\u003cspan address=\"10.1056/NEJMoa1801946\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCao W, Tang Q, Zeng J, Jin X, Zu L, Xu S. A Review of Biomarkers and Their Clinical Impact in Resected Early-Stage Non-Small-Cell Lung Cancer. Cancers (Basel). 2023;15:4561. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/cancers15184561\u003c/span\u003e\u003cspan address=\"10.3390/cancers15184561\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiddha P, Thummalapalli R, Quandt Z, Balaratnam K, Cardenas E, Falcon CJ, et al. Germline prediction of immune checkpoint inhibitor discontinuation for immune-related adverse events. J Immunother Cancer. 2025;13:e011273. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/jitc-2024-011273\u003c/span\u003e\u003cspan address=\"10.1136/jitc-2024-011273\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChin IS, Khan A, Olsson-Brown A, Papa S, Middleton G, Palles C. Germline genetic variation and predicting immune checkpoint inhibitor induced toxicity. npj Genom Med Nat Publishing Group. 2022;7:1\u0026ndash;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41525-022-00345-6\u003c/span\u003e\u003cspan address=\"10.1038/s41525-022-00345-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGusev A. Germline mechanisms of immunotherapy toxicities in the GWAS era. Immunol Rev. 2023;318:138\u0026ndash;56. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/imr.13253\u003c/span\u003e\u003cspan address=\"10.1111/imr.13253\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWeidhaas J, Marco N, Scheffler AW, Kalbasi A, Wilenius K, Rietdorf E, et al. Germline biomarkers predict toxicity to anti-PD1/PDL1 checkpoint therapy. J Immunother Cancer. 2022;10:e003625. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/jitc-2021-003625\u003c/span\u003e\u003cspan address=\"10.1136/jitc-2021-003625\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMonaco A, Pantaleo E, Amoroso N, Lacalamita A, Lo Giudice C, Fonzino A, et al. A primer on machine learning techniques for genomic applications. Comput Struct Biotechnol J. 2021;19:4345\u0026ndash;59. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.csbj.2021.07.021\u003c/span\u003e\u003cspan address=\"10.1016/j.csbj.2021.07.021\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAl-Tashi Q, Saad MB, Muneer A, Qureshi R, Mirjalili S, Sheshadri A, et al. Machine Learning Models for the Identification of Prognostic and Predictive Cancer Biomarkers: A Systematic Review. Int J Mol Sci. 2023;24:7781. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/ijms24097781\u003c/span\u003e\u003cspan address=\"10.3390/ijms24097781\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBhushan R, Rai S, Jangra P, Sonker P, Singh N, Lakshami R, et al. The role of artificial intelligence in predicting cancer immunotherapy response. Oral Oncol Rep. 2025;16:100767. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.oor.2025.100767\u003c/span\u003e\u003cspan address=\"10.1016/j.oor.2025.100767\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRakaee M, Tafavvoghi M, Ricciuti B, Alessi JV, Cortellini A, Citarella F, et al. Deep Learning Model for Predicting Immunotherapy Response in Advanced Non-Small Cell Lung Cancer. JAMA Oncol. 2025;11:109\u0026ndash;18. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jamaoncol.2024.5356\u003c/span\u003e\u003cspan address=\"10.1001/jamaoncol.2024.5356\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, Sargent D, Ford R, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer. 2009;45:228\u0026ndash;47. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ejca.2008.10.026\u003c/span\u003e\u003cspan address=\"10.1016/j.ejca.2008.10.026\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVan der Auwera GA, Carneiro MO, Hartl C, Poplin R, del Angel G, Levy-Moonshine A, et al. From FastQ data to high confidence variant calls: the Genome Analysis Toolkit best practices pipeline. Curr Protoc Bioinf. 2013;11:11. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/0471250953.bi1110s43\u003c/span\u003e\u003cspan address=\"10.1002/0471250953.bi1110s43\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen S, Zhou Y, Chen Y, Gu J. fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics. 2018;34:i884\u0026ndash;90. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/bioinformatics/bty560\u003c/span\u003e\u003cspan address=\"10.1093/bioinformatics/bty560\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcKenna A, Hanna M, Banks E, Sivachenko A, Cibulskis K, Kernytsky A, et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 2010;20:1297\u0026ndash;303. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1101/gr.107524.110\u003c/span\u003e\u003cspan address=\"10.1101/gr.107524.110\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMasica DL, Douville C, Tokheim C, Bhattacharya R, Kim R, Moad K, et al. CRAVAT 4: Cancer-Related Analysis of Variants Toolkit. Cancer Res. 2017;77:e35\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1158/0008-5472.CAN-17-0338\u003c/span\u003e\u003cspan address=\"10.1158/0008-5472.CAN-17-0338\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKarczewski KJ, Francioli LC, Tiao G, Cummings BB, Alf\u0026ouml;ldi J, Wang Q, et al. The mutational constraint spectrum quantified from variation in 141,456 humans. Nat Nat Publishing Group. 2020;581:434\u0026ndash;43. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41586-020-2308-7\u003c/span\u003e\u003cspan address=\"10.1038/s41586-020-2308-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRichards S, Aziz N, Bale S, Bick D, Das S, Gastier-Foster J, et al. Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. Genet Med Nat Publishing Group. 2015;17:405\u0026ndash;23. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/gim.2015.30\u003c/span\u003e\u003cspan address=\"10.1038/gim.2015.30\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGao J, Aksoy BA, Dogrusoz U, Dresdner G, Gross B, Sumer SO, et al. Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal. Sci Signal. 2013;6:pl1. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1126/scisignal.2004088\u003c/span\u003e\u003cspan address=\"10.1126/scisignal.2004088\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChao BN, Carrick DM, Filipski KK, Nelson SA. Overview of Research on Germline Genetic Variation in Immune Genes and Cancer Outcomes. Cancer Epidemiol Biomarkers Prev. 2022;31:495\u0026ndash;506. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1158/1055-9965.EPI-21-0583\u003c/span\u003e\u003cspan address=\"10.1158/1055-9965.EPI-21-0583\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYin J, Lu Y, Liu Y, Shi Q, Shi M, Zhu Z, et al. SIGLEC11 promotes M2 macrophage polarization through AKT-mTOR signaling and facilitates the progression of gastric cancer. J Immunother Cancer. 2025;13:e010162. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/jitc-2024-010162\u003c/span\u003e\u003cspan address=\"10.1136/jitc-2024-010162\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFeng B, Pan B, Huang J, Du Y, Wang X, Wu J, et al. PDE4D/cAMP/IL-23 axis determines the immunotherapy efficacy of lung adenocarcinoma via activating the IL-9 autocrine loop of cytotoxic T lymphocytes. Cancer Lett. 2023;565:216224. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.canlet.2023.216224\u003c/span\u003e\u003cspan address=\"10.1016/j.canlet.2023.216224\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRen H-L, Zhang S-H, Li P-Y. The multifaceted role of phosphodiesterase 4 in tumor: from tumorigenesis to immunotherapy. Front Immunol. 2025;16:1528932. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fimmu.2025.1528932\u003c/span\u003e\u003cspan address=\"10.3389/fimmu.2025.1528932\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUdagawa C, Horinouchi H, Shiraishi K, Kohno T, Okusaka T, Ueno H, et al. Whole genome sequencing to identify predictive markers for the risk of drug-induced interstitial lung disease. PLoS ONE. 2019;14:e0223371. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0223371\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0223371\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWright PK. Targeting vesicle trafficking: an important approach to cancer chemotherapy. Recent Pat Anticancer Drug Discov. 2008;3:137\u0026ndash;47. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2174/157489208784638730\u003c/span\u003e\u003cspan address=\"10.2174/157489208784638730\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbdul Aziz NA, Mokhtar NM, Harun R, Mollah MMH, Mohamed Rose I, Sagap I, et al. A 19-Gene expression signature as a predictor of survival in colorectal cancer. BMC Med Genomics. 2016;9:58. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12920-016-0218-1\u003c/span\u003e\u003cspan address=\"10.1186/s12920-016-0218-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZengin HY, Karabulut E. Biomarker detection using corrected degree of domesticity in hybrid social network feature selection for improving classifier performance. BMC Bioinformatics. 2023;24:407. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12859-023-05540-5\u003c/span\u003e\u003cspan address=\"10.1186/s12859-023-05540-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Z, Zhou Y, Takagi T, Song J, Tian Y-S, Shibuya T. Genetic algorithm-based feature selection with manifold learning for cancer classification using microarray data. BMC Bioinformatics. 2023;24:139. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12859-023-05267-3\u003c/span\u003e\u003cspan address=\"10.1186/s12859-023-05267-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHauskrecht M, Pelikan R, Valko M, Lyons-Weiler J. Feature Selection and Dimensionality Reduction in Genomics and Proteomics. In: Werner Dubitzky MG, Berrar D, editors. Fundamentals of Data Mining in Genomics and Proteomics. Springer; 2006. pp. 149\u0026ndash;72. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-0-387-47509-7\u003c/span\u003e\u003cspan address=\"10.1007/978-0-387-47509-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\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":true,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"journal-of-translational-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jtrm","sideBox":"Learn more about [Journal of Translational Medicine](http://translational-medicine.biomedcentral.com)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/jtrm/default.aspx","title":"Journal of Translational Medicine","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Non-small cell lung cancer, Germline variation, Immunotherapy biomarkers, Machine learning, Progression-free survival","lastPublishedDoi":"10.21203/rs.3.rs-8455214/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8455214/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003e Immune checkpoint inhibitors (ICIs) have transformed the treatment landscape of advanced non-small cell lung cancer (NSCLC). However, a substantial proportion of patients do not experience durable clinical benefits, and established tumour biomarkers, such as PD-L1 expression and tumour mutational burden, often show limited predictive value. The potential of inherited germline variants to predict immunotherapy outcomes in NSCLC remains a critical and underexplored area.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods:\u003c/strong\u003e We prospectively enrolled 117 patients with advanced NSCLC treated with ICI-based regimens at two centres in Spain. Germline whole-exome sequencing (WES) was performed on pretreatment blood samples. Exonic and intronic variants were annotated and integrated with comprehensive clinical data. We applied XGBoost and LASSO machine learning models to identify predictive germline variants and clinical features, and subsequently trained them to predict treatment response and progression-free survival (PFS). This approach produced a novel clinical–germline risk score, generating both a global model and a specific model for the lung adenocarcinoma (LUAD) histological subtype.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e XGBoost significantly outperformed penalised regression (LASSO), achieving a robust cross-validated area under the curve (AUC) of 0.845 for predicting treatment response in the validation cohort. Our models identified several novel germline loci that were significantly associated with immunotherapy outcomes. Variants in \u003cem\u003eSLC6A16\u003c/em\u003e, \u003cem\u003eSIGLEC11\u003c/em\u003e, \u003cem\u003ePDE4E\u003c/em\u003e, and \u003cem\u003eOR10H5\u003c/em\u003e were associated with reduced PFS, whereas variants in \u003cem\u003eCCZ1B\u003c/em\u003e and \u003cem\u003ePHLDB1\u003c/em\u003e were associated with extended PFS. Lymph node metastasis was confirmed as the sole independent clinical predictor of poor response (OR 2.07, P=0.008). A predictive algorithm that included these individual variables generated a clinical–germline risk score that successfully stratified patients into high- and low-risk groups with markedly different median PFS (low-risk: 18 months vs. high-risk: 7 months, log-rank P \u0026lt; 0.001), retaining discriminatory power across histological subgroups.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions:\u003c/strong\u003e The integration of specific germline variants as the output of advanced machine learning analysis of the exome with key clinical features provides accurate and novel predictive information for immunotherapy in NSCLC. This approach not only uncovers new genetic biomarkers but also supports the clinical adoption of composite risk scores for personalised precision immunotherapy, paving the way for improved patient selection and stratification.\u003c/p\u003e","manuscriptTitle":"Prospective Germline Exome and Machine Learning-Based Risk Score Identify Predictive and PrognosticBiomarkers of Immunotherapy Outcomes in Advanced Non-Small Cell Lung Cancer","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-20 09:46:31","doi":"10.21203/rs.3.rs-8455214/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"","date":"2026-02-10T09:04:29+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-01-16T05:36:57+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-01-14T04:05:48+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of Translational Medicine","date":"2026-01-08T03:38:35+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-translational-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jtrm","sideBox":"Learn more about [Journal of Translational Medicine](http://translational-medicine.biomedcentral.com)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/jtrm/default.aspx","title":"Journal of Translational Medicine","twitterHandle":"@BioMedCentral","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"8d4b2679-12e7-4223-b393-4e1116153c3b","owner":[],"postedDate":"January 20th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-04-04T02:19:09+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-20 09:46:31","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8455214","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8455214","identity":"rs-8455214","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}