Supervised classification of Preeclampsia clinical cases using datasets from MALDI-TOF-MS and machine learning tools

Supervised classification of Preeclampsia clinical cases using datasets from MALDI-TOF-MS and machine learning tools | 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 Supervised classification of Preeclampsia clinical cases using datasets from MALDI-TOF-MS and machine learning tools Luisa Amezquita, Laura Maca, Yuly Andrea Prada, Claudia Colmenares-Mejía, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6966953/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Preeclampsia is a pregnancy-induced disorder characterized by hypertension and high levels of proteinuria after 20 weeks of pregnancy. This condition significantly raises the risk of maternal-fetal death and increases the risk of vascular diseases after pregnancy. However, timely and unequivocal diagnosis in the first weeks allows access to appropriate medical follow-up. In this work, we performed a supervised classificatory analysis by machine learning using the protein profiles present in blood serum samples from 97 controls and 67 patient cases obtained by matrix-assisted laser ionization/desorption time-of-flight mass spectrometry. Results The protein profile was obtained from peptide mass fingerprinting of the samples by the Filter-Assisted Sample Preparation (FASP) protocol and subsequent acquisition of the mass spectra. The spectrum data analysis using machine learning algorithms demonstrated high performance in classifying cases and controls, with an overall accuracy of 88% and a sensitivity of 0.90 and 0.85 for predicting positive cases and negative controls, respectively. Conclusions These results highlight the reliability and versatility of analyzing and processing spectral data from mass spectrometry using artificial intelligence tools to study of preeclampsia. This study could pave the way for future applications in clinical diagnostics, offering new alternatives for improved patient outcomes. Preeclampsia Protein-fingerprinting Proteomics Mass spectrometry Machine learning Figures Figure 1 Figure 2 Figure 3 Background Preeclampsia is a hypertensive disorder of pregnancy that affects approximately 10% of pregnant women worldwide. It is associated with more than 50,000 maternal deaths and over 500,000 fetal deaths annually [1,2]. Preeclampsia is a multifactorial disease, posing challenges in clinical research due to its complex origins and molecular mechanisms [3–5]. A proportion of the deaths associated with pre-eclampsia can be avoided by timely detection and effective health care for women with hypertensive symptoms [6]. Therefore, several scientific approaches are currently emphasizing the need for and importance of identifying potential biomarkers of pre-eclampsia that can contribute to the early and reproducible identification of pregnant women at increased risk of developing preeclampsia [7,8]. In recent years, omics profiles in pregnancy have been investigated for different diseases, including type 1 diabetes and preeclampsia [9–11]. In this sense, high-resolution instruments and the most sensitive analytical techniques, such as MALDI-TOF-MS, have been used to analyze proteins and metabolites associated with these clinical outcomes [12–14].In the case of proteomics profiles, exhaustive and careful sample preparation steps are required, making it necessary to implement protocols for protein digestion to improve the resolution and reduce the sample complexity to facilitate the analysis of smaller fragments known as protein fingerprints by MALDI-TOF-MS [15,16]. Subsequently, analyzing mass spectral data using multivariate statistical methods has provided significant insight into the distinguishable characteristics of cases and controls in clinical studies for pre-eclampsia, using both urine and plasma samples [17,18]. However, in some cases, multivariate methods fail to classify groups with high precision due to the complexity of the biological sample [19,20]. Recently, artificial intelligence has been significantly helpful in identifying patterns that are undetectable by classical data analysis methods [21]. Applying machine learning (ML) algorithms to construct mathematical supervised models for classifying a sample within a group has enabled the development of disease detection methods [22,23]. The acquisition of protein profiles from samples for selecting training data and test data has become a valuable tool in medical research [24–26]. In this study, we aimed to develop an algorithm in the open-source programming language Python to classify patients with preeclampsia based on the proteomic profile of blood serum samples from cases and controls. Methods Reagents . All solutions were prepared in deionized water (resistive 18.2 MΩ.cm). Analytical grade Urea, Tris-HCl, HPLC grade acetonitrile (ACN), and trifluoroacetic acid (TFA) were purchased from Supelco Solutions (Merck, Darmstadt, Germany). Protein concentration was determined by Pierce BCA Protein Kit (Thermo Fisher Scientific, Massachusetts, USA). Sequencing grade trypsin, acetic acid (CH 3 COOH), analysis grade dithiothreitol (DTT), and iodoacetamide (IAA) were purchased from Fisher Bioreagents (Thermo Fisher Scientific, Massachusetts, USA). The matrix used for MS-MALDI was α-Cyano-4-hydroxycinnamic acid (CHCA), purchased from Sigma-Aldrich Solutions (Merck, Darmstadt, Germany). Vapreotide was used as a commercial peptide calibrant standard from Roche Diagnostics (Hoffmann-La Roche, Basel, Switzerland). Subjects and sample collection . GenPE was a prospective case-control study that recruited young women (< 26 years old) who were primigravid and previously healthy between 2000 and 2012 from various cities in Colombia. Cases corresponded to women at any gestational age with a confirmed diagnosis of preeclampsia (hypertension ≥ 140/90 mmHg in two separate measurements and proteinuria, ≥ 300 mg in 24 hours or ≥ 1 + dipstick reading in a random urine sample with no evidence of urinary tract infection after the 20th week of pregnancy). Controls were normotensive pregnant women without proteinuria and gestational age > 37 weeks recruited in the same hospital as the case [27,28]. Through probability sampling stratified by city and case/control status, serum samples were selected and retrieved from a biobank GenPE of patients who had given informed consent for future studies in preeclampsia. All serum samples were stored at -80 ºC until proteomic analysis. Sample protein total determination. Total protein was determined using the Pierce BCA Protein Kit according to Thermo Fisher's protocol. The proteins were reduced to the cuprous state. Ultraviolet-visible spectrophotometry (UV-Vis) analysis at 562 nm detected the blue complex formed. The absorbance intensity was proportional to the amount of protein in the sample. Filter-aided sample preparation method. Digestion by the FASP method was performed according to the protocol proposed by Jacek R. Wisniewski [29]. Then, 4 µL of blood serum samples were resuspended in 36 µL of UA solution (0.1 mol/L Tris-HCl at pH 8.5). Briefly, 2 µL of the samples were kept on a membrane filter with a molecular weight cut-off (MWCO) of 3000 Da. Consequently, 200 µL of UA was added onto the filter and centrifuged at 10000 g for 15 min at 10°C. After, 100 µL of DTT (0.05 mol/L in UA solution) was added above the filter, and the samples were incubated for 30 min at 37°C. The samples were centrifuged at 10000 g for 15 min at 10°C twice. Afterwards, the samples were incubated for 30 minutes with IAA solution (0.05 mol/L in UA solution). The excess of reagents was eliminated by two washes with 100 µL of UB (0.05 mol/L in buffer Tris-HCl at pH 8.5). Finally, 40 µL of UB was added to the amicon filter, and 2 µL of trypsin solution (2 µg-µL − 1 ) was added to each sample. The samples digestion was carried out at 37°C for 18 hours. Finally, the protein fragments were extracted from the amicon filter by centrifugation at 10000 g for 15 min with 100 µL of the UB solution twice. Then, the digested protein fragments in the eluted fraction were collected and concentrated using a vacuum concentrator (SpeedVac, Thermo Scientific) setup at 38ºC and stored at -40°C for subsequent analysis by MALDI-TOF-MS. Analysis by MALDI-TOF Mass Spectrometry . Each trypsinized sample was resuspended in 50 µL of TA30 (H 2 O/ACN 70:30; 0.01% TFA). After, an aliquot of 1 µL of the sample was diluted with TA30 at 20 µL of final volume and was deposited in a holder using a double-layer method. Briefly, 0.5 µL of CHCA was added to the spot, 0.5 µL of the samples, and 0.5 µL of CHCA [30,31]. Mass spectra were acquired in a range of 500 to 6000 Da or mass-charge ratio (m/z), with a laser power of 100%, at a frequency of 500 Hz. These characteristics were configured in a method established in the equipment software as Micro Bío Tools (MBT.par), which was used as the standard method for the MALDI-TOF-MS acquisition. Machine Learning Methods . The predictive model was developed in Python using Anaconda, an open-source software platform for artificial intelligence ( https://www.anaconda.com ), along with several libraries, including NumPy, Pandas, scikit-learn, and Matplotlib. The dataset was conformed to the ion m/z and the relative intensities from protein fragment spectra. We applied two learning approaches to build an unsupervised and a supervised model based on MALDI spectra for cases and controls. The dataset supporting these analyses is available in the GitHub repository ( https://github.com/YAPrada/MALDI-Data-Preeclampsia ) and summarized in Table S1 . For the supervised model, the data were stratified, with 80% allocated to the training phase and the remaining 20 per cent allocated to the testing phase. The spectra data were imported in .txt format to the Anaconda interface ( https://www.anaconda.com ) when the case (yes) and control (no) labels were designed (MALDI spectral data reservoir available on Supporting Information). Spectrum data was separated into intensity and m/z values and checked for no nulls for the baseline filled and intensities were normalized. The dataset was divided into 80% for training and 20% for test sets based on binary labels (1: case, 0: control). We used SVM, LR, RF, and XGBoost for supervised learning algorithms. The models were evaluated using various performance metrics, including accuracy, precision, sensitivity, and F1 score. Finally, the confusion matrix illustrates how the model classifies different classes. This step was focused on identifying features that significantly influenced supervised model predictions [32]. Results and Discussion The total protein content of the 164 blood serum samples was 28 and 160 µg-mL − 1 . Subsequently, mass spectra were obtained for 97 controls and 67 cases. However, as shown in Fig. 1, the visual comparison between cases and controls does not allow observing significant differences at first sight. A predictive model for the detection of preeclampsia based on the intensities corresponding to each m/z signal was developed using the mass spectra of peptide fragments obtained by MALDI-TOF-MS for cases and controls. The construction of this model involved both unsupervised and supervised learning methods, considering the visual complexity of identifying significant differences between the mass spectra of the two groups, as similarities were initially observed in the predominant signals around 2000 to 4500 Da. The superimposed spectra show no differences between the signals detected for cases and controls. However, differences in the relative intensity of some signals are observed in some regions of the spectrum. The similarity in the mass spectra of cases and controls is attributed to the complexity of the sample, where expression or overexpression of proteins associated with preeclampsia occurs at minimal concentrations [23]. Therefore, differentiating cases from controls by looking at the MALDI spectrum is imprecise and unreliable. Unsupervised models. The PCA model, an unsupervised learning approach for analyzing mass spectral intensities, aimed to reduce the dataset's dimensionality using eight principal components. The PCA results evidence this ability to capture the inherent variability of the data significantly [33]. Figure 2–3 showed that the dimensionality reduction was successfully achieved, but effective discrimination between the two groups was not achieved with a sum of explained variance of 0.81. Similarly, the pairs plot, generated from the dimensional reduction, seeks to represent the relationship between pairs of variables within the reduced set of variability, showing an unsatisfactory differentiation between cases and controls. However, an in-depth spectral data analysis under a supervised model was required to detect differences and find patterns or anomalies between the two groups. Supervised models. For the supervised models, we calculated accuracy as a global metric for each model; as shown in Table 1 , the SVM performed better according to accuracy with a value of 0.88, followed by LR and XGBoost with 0.78 and 0.75 for RF. Table 1 Metrics for the classification by classes (cases or controls) for the four supervised algorithms Metric Precision Recall F1-score Support class Model 0 1 0 1 0 1 0 1 SVM 0.95 0.79 0.86 0.92 0.90 0.85 21 12 LR 0.79 0.21 0.58 0.43 0.67 0.29 26 7 RF 0.76 0.75 0.90 0.50 0.83 0.60 21 12 XGBoost 0.82 0.73 0.86 0.67 0.84 0.70 21 12 According to the results, the lack of apparent separation between the groups indicates the need to explore complementary approaches to improve effective discrimination between the two groups. Therefore, supervised learning methods were implemented in this prospective study (Supplementary Information, Table S1 ). These learning methods are based on dividing the dataset into training and test data, for which 4 models were used (SVM, LR, RF, and XGBoost). The SVM model performed better in global metrics, achieving an accuracy of 0.88 compared to the other models. In this sense, the LR algorithm showed an inferior performance in the differentiation of cases and controls despite having an excellent overall metric. On the other hand, SVM presented a better accuracy than the different models, which reflects the percentage of correct predictions over the total predictions, clarifying that it is within the test percentage. Furthermore, sensitivity represents the ability of the model to correctly identify cases (positive) and controls (negative). In this case, the best results by class are divided into two models: SVM with a value of 0.92 for cases (1) and RF with a value of 0.90 for controls (0). Finally, the F1-score combines the two previous metrics to choose the best model, which in this case was SVM, which obtained the best performance in differentiating women with and without preeclampsia in a test dataset of 20 percent. However, it is crucial to consider detailed evaluation metrics per class due to their imbalance (97 controls, 67 cases), as shown in Table 1 , when calculating the complete values for all metrics. Although, despite having a good performance in this global metric, limitations in the model may be present, so one should not rely on a single metric to measure the performance of the predictive model, so a thorough analysis of other more detailed classification metrics by class may provide a better understanding of the model performance, these were accuracy, recall, and F1-Score. On the other hand, the confusion matrix allows an accurate visual representation of each classification model for the test data. Identifying true and false cases, as well as true and false controls. In the test set, specifically, the 20 percent reserved, it was evident that the RL was the model with the highest error rate during training, as is shown in Figure S1 in the supplementary information. There is a general tendency for the evaluation models to perform better at predicting cases than controls. However, it is essential to highlight that an inferior performance was observed in predicting false cases and controls. These evaluation metrics enabled the establishment of the most effective classification model to distinguish between patients with and without preeclampsia, using blood serum samples that had been previously digested and characterized by MALDI-TOF-MS. The SVM algorithm achieved an accuracy of 0.88, results that are consistent with previous research. For instance, a review of the most widely implemented models for preeclampsia diagnosis emphasized the importance of recall as an evaluation metric in clinical studies, especially when faced with minimal differences between cases and controls [34,35]. Other research reported a high efficiency of RF, followed by SVM and RL models in analyzing clinical data from women with preeclampsia. The SVM model achieved an accuracy of 0.68 and a precision of 0.51, while the RF model had a precision of 0.86 [36]. Compared to other research using intact protein samples, where outstanding results were obtained for binary classification, the high versatility of the data from mass spectrometry in developing predictive mathematical models is evident, which supports the suitability of the methodology used in this study for digested protein samples [37]. In this study, differences in the intensity of the mass spectra between the case and control groups are observed, enabling the identification of distinct patterns between the sample groups of preeclampsia. The SVM model demonstrated superior performance in the cohort of pregnant women in the study. This result contrasts with previous findings where the differences between cases and controls for preeclampsia were minimal, and the SVM performed poorly in predicting [38]. This discrepancy may be due to the complexity of the disease, variations in data analysis methods, and the number of datasets. Limitations and perspectives. This study presents a novel approach to the application of machine learning models for the clinical classification of preeclampsia cases. Despite this approach's inherent limitations, such as the need for large data sets, these methods can complement traditional techniques by providing deeper insight into the underlying patterns in mass spectra. This modelling with a big data set improved predictive ability and, ultimately, improved clinical care for pregnant women at risk for preeclampsia. Furthermore, it encourages research in the clinical field by relying on such computational tools and spectrometric data. Conclusions Through this study, we implemented a preparatory method based on the FASP protocol to obtain the peptide fingerprint of blood serum samples from preeclampsia cases and controls, thereby generating high-quality mass spectra. We then formulated a predictive model based on these data using machine learning tools. The algorithm based on support vector machines was the one that obtained the highest accuracy (0.88) and recall (0.92) to predict the true cases as evident in the results of the metrics expressed in the confusion matrix. These results represent a significant advance in disease detection and pave the way for future studies in the timely diagnosis of conditions such as Preeclampsia, supported by machine learning data analysis. In addition, this research demonstrates the versatility of the FASP method in protein sample preparation and data analysis using supervised classification algorithms, which can be applied to the study of other complex multifactorial diseases. Declarations Ethics approval and consent to participate. This study was conducted using samples collected in the GenPE project, which established a biobank of information and human biological material (maternal and neonatal) through the active recruitment of cases and controls over a 12-year period (2000-2012). During this recruitment process and before entering the study, written informed consent was obtained from the pregnant women to participate in the research and to store a blood sample and its derivatives for analysis related to the study of preeclampsia. Data Availability. Labels for case and control samples are summarized in Table S1 on the Information Support used in this study, with the MALDI spectral data reservoir are publicly available on: https://github.com/YAPrada/MALDI-Data-Preeclampsia. And we present in the Information Supporting Figure S1. Confusion matrices for all machine learning algorithms for the classification of Preeclampsia cases. Code Availability. The codes are available to public access as a notebook on: https://github.com/Lauracmaca/MODELO-PE-MALDI-TOF/blob/main/Support%20Information%20S1%205-7-2024%20.ipynb. Competing interests. The authors declare no competing interests. Funding. This work was supported by the OEI Program ‘Women, Science and Equity’ through the Ministry of Science, Technology and Innovation (MinCiencias), Colombia. Authors’ Contributions. LA † and LM † performed the experimental probes, training/testing datasets, and wrote the manuscript. YAP, PB, and CC supervised the investigation, data interpretation, and manuscript review and editing. NC, DC, and EMO designed this study and reviewed the manuscript. All authors have approved the final version of the manuscript. † These authors contributed equally Acknowledgments . We thank the Mass Spectrometry Laboratory of Parque Tecnológico Guatiguará at Universidad Industrial de Santander for allowing us to use the mass spectrometer for the spectral analysis. References Sammour MB, El-Kabarity H, Fawzy MM, Schindler a. E. Prevention and treatment of pre-eclampsia and eclampsia. Journal of Steroid Biochemistry & Molecular Biology. 2011. Navajas R, Corrales F, Paradela A. Quantitative proteomics-based analyses performed on pre-eclampsia samples in the 2004–2020 period: a systematic review. Clin Proteomics [Internet]. 2021;18:1–12. Available from: https://doi.org/10.1186/s12014-021-09313-1 Lu HQ, Hu R. Lasting effects of intrauterine exposure to preeclampsia on offspring and the underlying mechanism. AJP Rep. 2019;9:E275–91. Bellamy L, Casas JP, Hingorani AD, Williams DJ. Pre-eclampsia and risk of cardiovascular disease and cancer in later life: Systematic review and meta-analysis. Br Med J. 2007;335:974–7. Atluri N, Beyuo TK, Oppong SA, Moyer CA, Lawrence ER. Challenges to diagnosing and managing preeclampsia in a low-resource setting: A qualitative study of obstetric provider perspectives from Ghana. PLOS Global Public Health [Internet]. 2023;3:e0001790. Available from: http://dx.doi.org/10.1371/journal.pgph.0001790 Opichka MA, Rappelt MW, Gutterman DD, Grobe JL, McIntosh JJ. Review vascular dysfunction in preeclampsia. Cells. 2021;10. Sahai K, Saraswathy S, Yadav TP, Arora D, Krishnan M. Pre-eclampsia: Molecular events to biomarkers. Med J Armed Forces India [Internet]. 2017;73:167–74. Available from: http://dx.doi.org/10.1016/j.mjafi.2016.09.001 Nobakht M. Gh BF. Application of metabolomics to preeclampsia diagnosis. Syst Biol Reprod Med [Internet]. 2018;64:324–39. Available from: https://doi.org/10.1080/19396368.2018.1482968 Tarca AL, Romero R, Benshalom-Tirosh N, Than NG, Gudicha DW, Done B, et al. The prediction of early preeclampsia: Results from a longitudinal proteomics study. PLoS One. 2019. Ghaemi MS, Tarca AL, Romero R, Stanley N, Fallahzadeh R, Tanada A, et al. Proteomic signatures predict preeclampsia in individual cohorts but not across cohorts–implications for clinical biomarker studies. Journal of Maternal-Fetal and Neonatal Medicine [Internet]. 2022;35:5621–8. Available from: https://doi.org/10.1080/14767058.2021.1888915 Than NG, Romero R, Györffy D, Posta M, Bhatti G, Done B, et al. Molecular subclasses of preeclampsia characterized by a longitudinal maternal proteomics study: Distinct biomarkers, disease pathways and options for prevention. J Perinat Med. 2023;51:51–68. Banach P, Dereziński P, Matuszewska E, Matysiak J, Bochyński H, Kokot ZJ, et al. MALDI-TOF-MS analysis in the identification of urine proteomic patterns of gestational trophoblastic disease. Metabolites. 2019;9. Rozanova S, Barkovits K, Nikolov M, Schmidt C, Urlaub H, Marcus K. Quantitative Mass Spectrometry-Based Proteomics: An Overview. Methods in Molecular Biology. 2021. Zaikin VG, Borisov RS. Mass Spectrometry as a Crucial Analytical Basis for Omics Sciences. Journal of Analytical Chemistry. 2021;76:1567–87. Darie-Ion L, Whitham D, Jayathirtha M, Rai Y, Neagu AN, Darie CC, et al. Applications of MALDI-MS/MS-Based Proteomics in Biomedical Research. Molecules. 2022;27:1–23. Klein J, Buffin-Meyer B, Mullen W, Carty DM, Delles C, Vlahou A, et al. Clinical proteomics in obstetrics and neonatology. Expert Rev Proteomics. 2014;11:75–89. Chen H, Aneman I, Nikolic V, Karadzov Orlic N, Mikovic Z, Stefanovic M, et al. Maternal plasma proteome profiling of biomarkers and pathogenic mechanisms of early-onset and late-onset preeclampsia. Sci Rep [Internet]. 2022;12:1–13. Available from: https://doi.org/10.1038/s41598-022-20658-x Bujold E, Fillion A, Roux-Dalvai F, Scott-Boyer MP, Giguère Y, Forest JC, et al. Proteomic analysis of maternal urine for the early detection of preeclampsia and fetal growth restriction. J Clin Med. 2021;10. Davodabadi A, Daneshian B, Saati S, Razavyan S. Mathematical model and artificial intelligence for diagnosis of Alzheimer’s disease. Eur Phys J Plus [Internet]. 2023;138. Available from: https://doi.org/10.1140/epjp/s13360-023-04128-5 Sedaghati F, Gleason RL. A mathematical model of vascular and hemodynamics changes in early and late forms of preeclampsia. Physiol Rep. 2023;11:1–21. Mészáros B, Kukor Z, Valent S. Recent Advances in the Prevention and Screening of Preeclampsia. J Clin Med. 2023;12. Carreno JF, Qiu P. Feature selection algorithms for predicting preeclampsia: A comparative approach. Proceedings - 2020 IEEE International Conference on Bioinformatics and Biomedicine, BIBM 2020. 2020;2626–31. Liu M, Yang X, Chen G, Ding Y, Shi M, Sun L, et al. Development of a prediction model on preeclampsia using machine learning-based method: a retrospective cohort study in China. Front Physiol. 2022;13:1–9. Sarkar D, Saha S. Machine-learning techniques for the prediction of protein–protein interactions. J Biosci. 2019;44. Aljameel SS, Alzahrani M, Almusharraf R, Altukhais M, Alshaia S, Sahlouli H, et al. Prediction of Preeclampsia Using Machine Learning and Deep Learning Models: A Review. Big Data and Cognitive Computing. 2023;7. Thompson J, Everhart Nunn SL, Sarkar S, Clayton B. Diagnostic Screening of Bovine Mastitis Using MALDI-TOF MS Direct-Spotting of Milk and Machine Learning. Vet Sci. 2023;10. Serrano NC, Guio-Mahecha E, Quintero-Lesmes DC, Becerra-Bayona S, Paez MC, Beltran M, et al. Lipid profile, plasma apolipoproteins, and pre-eclampsia risk in the GenPE case-control study. Atherosclerosis. 2018;276:189–94. Colmenares-Mejia CC, Quintero-Lesmes DC, Bautista-Niño PK, Guío E, Paez MC, Beltrán M, et al. Uric acid and risk of pre-eclampsia: results from a large case–control study and meta-analysis of prospective studies. Sci Rep [Internet]. 2023;13:1–10. Available from: https://doi.org/10.1038/s41598-023-29651-4 Wiśniewski JR. Filter-aided sample preparation for proteome analysis. Methods in Molecular Biology. 2018;1841:3–10. Daltonics B. Bruker Guide to MALDI Sample Preparation [Internet]. MALDI Preparation Protocols. 2012. Available from: www.care-bdal.com Dave K a, Headlam MJ, Wallis TP, Gorman JJ. Preparation and analysis of proteins and peptides using MALDI TOF/TOF mass spectrometry. Current protocols in protein science / editorial board, John E Coligan . [et al]. 2011;Chapter 16:Unit 16.13. Liu Y, Wu R, Yang A. Research on Medical Problems Based on Mathematical Models. Mathematics. 2023;11. Ke M, Shen H, Wang L, Luo S, Lin L, Yang J, et al. Sample Preparation, Analysis and Practical Applications for mass spectrometry-based proteomics [Internet]. Adv Exp Med Biol. 2016. Available from: http://www.springer.com/gp/book/9783319414461%0Ahttp://link.springer.com/10.1007/978-3-319-41448-5 Schauberger G, Klug SJ, Berger M. Random forests for the analysis of matched case–control studies. BMC Bioinformatics. 2024;25. Huang S, Nianguang CAI, Penzuti Pacheco P, Narandes S, Wang Y, Wayne XU. Applications of support vector machine (SVM) learning in cancer genomics. Cancer Genomics Proteomics. International Institute of Anticancer Research; 2018. p. 41–51. Zhang D, Hu Y, Guo W, Song Y, Yang L, Yang S, et al. Mendelian randomization study reveals a causal relationship between rheumatoid arthritis and risk for pre-eclampsia. Front Immunol. 2022;13:1–8. Yang J, Gao Z, Ren X, Sheng J, Xu P, Chang C, et al. DeepDigest: Prediction of Protein Proteolytic Digestion with Deep Learning. Anal Chem. 2021;93:6094–103. Jhee JH, Lee S, Park Y, Lee SE, Kim YA, Kang SW, et al. Prediction model development of late-onset preeclampsia using machine learning-based methods. PLoS One. 2019;14:1–12. Additional Declarations No competing interests reported. Supplementary Files SISupervisedclassificationofpreeclampsiaclinicalBioData.docx Cite Share Download PDF Status: Posted Version 1 posted 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-6966953","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":482905045,"identity":"540b8360-8b05-4168-9f3e-3efd6ee24929","order_by":0,"name":"Luisa Amezquita","email":"","orcid":"","institution":"Universidad Pedagógica y Tecnológica de Colombia (UPTC)","correspondingAuthor":false,"prefix":"","firstName":"Luisa","middleName":"","lastName":"Amezquita","suffix":""},{"id":482905046,"identity":"01de1f91-e1c3-4751-b377-b52e4f938eef","order_by":1,"name":"Laura Maca","email":"","orcid":"","institution":"Universidad del Cauca","correspondingAuthor":false,"prefix":"","firstName":"Laura","middleName":"","lastName":"Maca","suffix":""},{"id":482905047,"identity":"2d232179-daba-4c51-bd41-f334eff5be6f","order_by":2,"name":"Yuly Andrea Prada","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABDklEQVRIiWNgGAWjYFAC5gYIfYD5AFwMxGJswKkFJnWALYFkLTwGWMUxgMGNxDYJxh335PmOn/n44eOee3L8DDyGB38w2MhuOMD88AFOLWeKDWeeyd0sOeNZsbFkA4/BYR6GNOMNB9iMDXBqaUtg3HAgdxszz4GExA1AFx5mYDgMYrBJ4NFiv+H8m2cgLfUgLUCH/SeoJXHDjRw2kJYEA6CWAzwMB3BqkTzzsNki8UxC8swbz4wlZxxIMJzZzFZwmMcg2XjmYex+4TuefPDGxx0Jtn3nkx9++HAgQZ6fvXnzxx8VdrJ9x5uxhpjCAQYWicQGZCFmsINhDEwg38DA/AFPRI+CUTAKRsEoYGAAAOBkbtQyqNJ/AAAAAElFTkSuQmCC","orcid":"","institution":"Universidad Industrial de Santander. Laboratorio de Espectroscopia Atómica y Molecular (LEAM), Parque Tecnológico Guatiguara. Piedecuesta","correspondingAuthor":true,"prefix":"","firstName":"Yuly","middleName":"Andrea","lastName":"Prada","suffix":""},{"id":482905048,"identity":"91296199-b889-4d7e-806e-8c5fe155bc8a","order_by":3,"name":"Claudia Colmenares-Mejía","email":"","orcid":"","institution":"Fundación Cardiovascular de Colombia (FCV), Hospital Internacional de Colombia","correspondingAuthor":false,"prefix":"","firstName":"Claudia","middleName":"","lastName":"Colmenares-Mejía","suffix":""},{"id":482905049,"identity":"431d8b38-f877-4f3b-acdf-fd02e92f9a84","order_by":4,"name":"Doris Quintero-Lesmes","email":"","orcid":"","institution":"Fundación Cardiovascular de Colombia (FCV), Hospital Internacional de Colombia","correspondingAuthor":false,"prefix":"","firstName":"Doris","middleName":"","lastName":"Quintero-Lesmes","suffix":""},{"id":482905050,"identity":"a5c0f1fd-a0be-489b-adc5-f137f758e8a0","order_by":5,"name":"Paula Bautista-Niño","email":"","orcid":"","institution":"Fundación Cardiovascular de Colombia (FCV), Hospital Internacional de Colombia","correspondingAuthor":false,"prefix":"","firstName":"Paula","middleName":"","lastName":"Bautista-Niño","suffix":""},{"id":482905051,"identity":"ad790acb-a447-4d97-82a6-3b44f4d56f43","order_by":6,"name":"Norma Serrano","email":"","orcid":"","institution":"Fundación Cardiovascular de Colombia (FCV), Hospital Internacional de Colombia","correspondingAuthor":false,"prefix":"","firstName":"Norma","middleName":"","lastName":"Serrano","suffix":""},{"id":482905052,"identity":"77aba11f-56f5-4cab-87d7-79beb343ce94","order_by":7,"name":"Enrique Mejía-Ospino","email":"","orcid":"","institution":"Universidad Industrial de Santander. Laboratorio de Espectroscopia Atómica y Molecular (LEAM), Parque Tecnológico Guatiguara. Piedecuesta","correspondingAuthor":false,"prefix":"","firstName":"Enrique","middleName":"","lastName":"Mejía-Ospino","suffix":""}],"badges":[],"createdAt":"2025-06-24 14:53:05","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6966953/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6966953/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":86655828,"identity":"ef3a597c-9c2a-4e8e-b565-bc15d0f4b3ec","added_by":"auto","created_at":"2025-07-14 10:15:40","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":145554,"visible":true,"origin":"","legend":"\u003cp\u003eMALDI-TOF mass spectra of cases and controls and superposition spectra of both\u003c/p\u003e","description":"","filename":"Figura1.spectra.png","url":"https://assets-eu.researchsquare.com/files/rs-6966953/v1/862802ed872e412becbe5ea7.png"},{"id":86657730,"identity":"b95eeebb-dd68-494a-9cda-240c1819f244","added_by":"auto","created_at":"2025-07-14 10:23:40","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":143029,"visible":true,"origin":"","legend":"\u003cp\u003ePCA for cases and controls of serum samples in this study. A two-dimensional matrix or array was generated where the data for each spectrum was presented in terms of relative abundance and designated by sample number.\u003c/p\u003e","description":"","filename":"Figura2.PCA.png","url":"https://assets-eu.researchsquare.com/files/rs-6966953/v1/6d6feb78758ce484b2d47c0a.png"},{"id":86657728,"identity":"63c632a7-9dce-4b9c-ac34-e1f3fde3e4df","added_by":"auto","created_at":"2025-07-14 10:23:40","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":494886,"visible":true,"origin":"","legend":"\u003cp\u003eDimensionality reduction representation of the data for PCA. Pairwise diagram of the eight principal components and their respective relationships.\u003c/p\u003e","description":"","filename":"Figura3.PairDiagram.png","url":"https://assets-eu.researchsquare.com/files/rs-6966953/v1/7256b9d19bf34135779a2dc6.png"},{"id":96708109,"identity":"ca63e8bc-0fe2-4a8c-9a4e-95ac6ee725bd","added_by":"auto","created_at":"2025-11-25 09:56:54","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1360532,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6966953/v1/4c901ff9-af20-491f-a9a1-0013b62ac21d.pdf"},{"id":86655825,"identity":"96b6714e-add6-4426-82b2-5be151cf2b0f","added_by":"auto","created_at":"2025-07-14 10:15:40","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":24723,"visible":true,"origin":"","legend":"","description":"","filename":"SISupervisedclassificationofpreeclampsiaclinicalBioData.docx","url":"https://assets-eu.researchsquare.com/files/rs-6966953/v1/39c59d23064bf1f0e8f66686.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eSupervised classification of Preeclampsia clinical cases using datasets from MALDI-TOF-MS and machine learning tools\u003c/p\u003e","fulltext":[{"header":"Background","content":"\u003cp\u003ePreeclampsia is a hypertensive disorder of pregnancy that affects approximately 10% of pregnant women worldwide. It is associated with more than 50,000 maternal deaths and over 500,000 fetal deaths annually [1,2]. Preeclampsia is a multifactorial disease, posing challenges in clinical research due to its complex origins and molecular mechanisms [3\u0026ndash;5]. A proportion of the deaths associated with pre-eclampsia can be avoided by timely detection and effective health care for women with hypertensive symptoms [6]. Therefore, several scientific approaches are currently emphasizing the need for and importance of identifying potential biomarkers of pre-eclampsia that can contribute to the early and reproducible identification of pregnant women at increased risk of developing preeclampsia [7,8].\u003c/p\u003e\u003cp\u003eIn recent years, omics profiles in pregnancy have been investigated for different diseases, including type 1 diabetes and preeclampsia [9\u0026ndash;11]. In this sense, high-resolution instruments and the most sensitive analytical techniques, such as MALDI-TOF-MS, have been used to analyze proteins and metabolites associated with these clinical outcomes [12\u0026ndash;14].In the case of proteomics profiles, exhaustive and careful sample preparation steps are required, making it necessary to implement protocols for protein digestion to improve the resolution and reduce the sample complexity to facilitate the analysis of smaller fragments known as protein fingerprints by MALDI-TOF-MS [15,16].\u003c/p\u003e\u003cp\u003eSubsequently, analyzing mass spectral data using multivariate statistical methods has provided significant insight into the distinguishable characteristics of cases and controls in clinical studies for pre-eclampsia, using both urine and plasma samples [17,18].\u003c/p\u003e\u003cp\u003eHowever, in some cases, multivariate methods fail to classify groups with high precision due to the complexity of the biological sample [19,20]. Recently, artificial intelligence has been significantly helpful in identifying patterns that are undetectable by classical data analysis methods [21]. Applying machine learning (ML) algorithms to construct mathematical supervised models for classifying a sample within a group has enabled the development of disease detection methods [22,23]. The acquisition of protein profiles from samples for selecting training data and test data has become a valuable tool in medical research [24\u0026ndash;26]. In this study, we aimed to develop an algorithm in the open-source programming language Python to classify patients with preeclampsia based on the proteomic profile of blood serum samples from cases and controls.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cb\u003eReagents\u003c/b\u003e. All solutions were prepared in deionized water (resistive 18.2 MΩ.cm). Analytical grade Urea, Tris-HCl, HPLC grade acetonitrile (ACN), and trifluoroacetic acid (TFA) were purchased from Supelco Solutions (Merck, Darmstadt, Germany). Protein concentration was determined by Pierce BCA Protein Kit (Thermo Fisher Scientific, Massachusetts, USA). Sequencing grade trypsin, acetic acid (CH\u003csub\u003e3\u003c/sub\u003eCOOH), analysis grade dithiothreitol (DTT), and iodoacetamide (IAA) were purchased from Fisher Bioreagents (Thermo Fisher Scientific, Massachusetts, USA). The matrix used for MS-MALDI was α-Cyano-4-hydroxycinnamic acid (CHCA), purchased from Sigma-Aldrich Solutions (Merck, Darmstadt, Germany). Vapreotide was used as a commercial peptide calibrant standard from Roche Diagnostics (Hoffmann-La Roche, Basel, Switzerland).\u003c/p\u003e\u003cp\u003e\u003cb\u003eSubjects and sample collection\u003c/b\u003e. GenPE was a prospective case-control study that recruited young women (\u0026lt;\u0026thinsp;26 years old) who were primigravid and previously healthy between 2000 and 2012 from various cities in Colombia. Cases corresponded to women at any gestational age with a confirmed diagnosis of preeclampsia (hypertension\u0026thinsp;\u0026ge;\u0026thinsp;140/90 mmHg in two separate measurements and proteinuria, \u0026ge; 300 mg in 24 hours or \u0026ge;\u0026thinsp;1\u0026thinsp;+\u0026thinsp;dipstick reading in a random urine sample with no evidence of urinary tract infection after the 20th week of pregnancy). Controls were normotensive pregnant women without proteinuria and gestational age\u0026thinsp;\u0026gt;\u0026thinsp;37 weeks recruited in the same hospital as the case [27,28]. Through probability sampling stratified by city and case/control status, serum samples were selected and retrieved from a biobank GenPE of patients who had given informed consent for future studies in preeclampsia. All serum samples were stored at -80 \u0026ordm;C until proteomic analysis.\u003c/p\u003e\u003cp\u003e\u003cb\u003eSample protein total determination.\u003c/b\u003e Total protein was determined using the Pierce BCA Protein Kit according to Thermo Fisher's protocol. The proteins were reduced to the cuprous state. Ultraviolet-visible spectrophotometry (UV-Vis) analysis at 562 nm detected the blue complex formed. The absorbance intensity was proportional to the amount of protein in the sample.\u003c/p\u003e\u003cp\u003e\u003cb\u003eFilter-aided sample preparation method.\u003c/b\u003e Digestion by the FASP method was performed according to the protocol proposed by Jacek R. Wisniewski [29]. Then, 4 \u0026micro;L of blood serum samples were resuspended in 36 \u0026micro;L of UA solution (0.1 mol/L Tris-HCl at pH 8.5). Briefly, 2 \u0026micro;L of the samples were kept on a membrane filter with a molecular weight cut-off (MWCO) of 3000 Da. Consequently, 200 \u0026micro;L of UA was added onto the filter and centrifuged at 10000 g for 15 min at 10\u0026deg;C. After, 100 \u0026micro;L of DTT (0.05 mol/L in UA solution) was added above the filter, and the samples were incubated for 30 min at 37\u0026deg;C. The samples were centrifuged at 10000 g for 15 min at 10\u0026deg;C twice. Afterwards, the samples were incubated for 30 minutes with IAA solution (0.05 mol/L in UA solution). The excess of reagents was eliminated by two washes with 100 \u0026micro;L of UB (0.05 mol/L in buffer Tris-HCl at pH 8.5). Finally, 40 \u0026micro;L of UB was added to the amicon filter, and 2 \u0026micro;L of trypsin solution (2 \u0026micro;g-\u0026micro;L\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e) was added to each sample. The samples digestion was carried out at 37\u0026deg;C for 18 hours. Finally, the protein fragments were extracted from the amicon filter by centrifugation at 10000 g for 15 min with 100 \u0026micro;L of the UB solution twice. Then, the digested protein fragments in the eluted fraction were collected and concentrated using a vacuum concentrator (SpeedVac, Thermo Scientific) setup at 38\u0026ordm;C and stored at -40\u0026deg;C for subsequent analysis by MALDI-TOF-MS.\u003c/p\u003e\u003cp\u003e\u003cb\u003eAnalysis by MALDI-TOF Mass Spectrometry\u003c/b\u003e. Each trypsinized sample was resuspended in 50 \u0026micro;L of TA30 (H\u003csub\u003e2\u003c/sub\u003eO/ACN 70:30; 0.01% TFA). After, an aliquot of 1 \u0026micro;L of the sample was diluted with TA30 at 20 \u0026micro;L of final volume and was deposited in a holder using a double-layer method. Briefly, 0.5 \u0026micro;L of CHCA was added to the spot, 0.5 \u0026micro;L of the samples, and 0.5 \u0026micro;L of CHCA [30,31]. Mass spectra were acquired in a range of 500 to 6000 Da or mass-charge ratio (m/z), with a laser power of 100%, at a frequency of 500 Hz. These characteristics were configured in a method established in the equipment software as Micro B\u0026iacute;o Tools (MBT.par), which was used as the standard method for the MALDI-TOF-MS acquisition.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMachine Learning Methods\u003c/b\u003e. The predictive model was developed in Python using Anaconda, an open-source software platform for artificial intelligence (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.anaconda.com\u003c/span\u003e\u003cspan address=\"https://www.anaconda.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), along with several libraries, including NumPy, Pandas, scikit-learn, and Matplotlib. The dataset was conformed to the ion m/z and the relative intensities from protein fragment spectra. We applied two learning approaches to build an unsupervised and a supervised model based on MALDI spectra for cases and controls. The dataset supporting these analyses is available in the GitHub repository (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/YAPrada/MALDI-Data-Preeclampsia\u003c/span\u003e\u003cspan address=\"https://github.com/YAPrada/MALDI-Data-Preeclampsia\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and summarized in Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e. For the supervised model, the data were stratified, with 80% allocated to the training phase and the remaining 20 per cent allocated to the testing phase.\u003c/p\u003e\u003cp\u003eThe spectra data were imported in .txt format to the Anaconda interface (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.anaconda.com\u003c/span\u003e\u003cspan address=\"https://www.anaconda.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) when the case (yes) and control (no) labels were designed (MALDI spectral data reservoir available on Supporting Information). Spectrum data was separated into intensity and m/z values and checked for no nulls for the baseline filled and intensities were normalized. The dataset was divided into 80% for training and 20% for test sets based on binary labels (1: case, 0: control). We used SVM, LR, RF, and XGBoost for supervised learning algorithms. The models were evaluated using various performance metrics, including accuracy, precision, sensitivity, and F1 score. Finally, the confusion matrix illustrates how the model classifies different classes. This step was focused on identifying features that significantly influenced supervised model predictions [32].\u003c/p\u003e"},{"header":"Results and Discussion","content":"\u003cp\u003eThe total protein content of the 164 blood serum samples was 28 and 160 \u0026micro;g-mL\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e. Subsequently, mass spectra were obtained for 97 controls and 67 cases. However, as shown in Fig.\u0026nbsp;1, the visual comparison between cases and controls does not allow observing significant differences at first sight.\u003c/p\u003e\u003cp\u003eA predictive model for the detection of preeclampsia based on the intensities corresponding to each m/z signal was developed using the mass spectra of peptide fragments obtained by MALDI-TOF-MS for cases and controls. The construction of this model involved both unsupervised and supervised learning methods, considering the visual complexity of identifying significant differences between the mass spectra of the two groups, as similarities were initially observed in the predominant signals around 2000 to 4500 Da. The superimposed spectra show no differences between the signals detected for cases and controls. However, differences in the relative intensity of some signals are observed in some regions of the spectrum. The similarity in the mass spectra of cases and controls is attributed to the complexity of the sample, where expression or overexpression of proteins associated with preeclampsia occurs at minimal concentrations [23]. Therefore, differentiating cases from controls by looking at the MALDI spectrum is imprecise and unreliable.\u003c/p\u003e\u003cp\u003e\u003cb\u003eUnsupervised models.\u003c/b\u003e The PCA model, an unsupervised learning approach for analyzing mass spectral intensities, aimed to reduce the dataset's dimensionality using eight principal components. The PCA results evidence this ability to capture the inherent variability of the data significantly [33]. Figure\u0026nbsp;2\u0026ndash;3 showed that the dimensionality reduction was successfully achieved, but effective discrimination between the two groups was not achieved with a sum of explained variance of 0.81.\u003c/p\u003e\u003cp\u003eSimilarly, the pairs plot, generated from the dimensional reduction, seeks to represent the relationship between pairs of variables within the reduced set of variability, showing an unsatisfactory differentiation between cases and controls. However, an in-depth spectral data analysis under a supervised model was required to detect differences and find patterns or anomalies between the two groups.\u003c/p\u003e\u003cp\u003e\u003cb\u003eSupervised models.\u003c/b\u003e For the supervised models, we calculated accuracy as a global metric for each model; as shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the SVM performed better according to accuracy with a value of 0.88, followed by LR and XGBoost with 0.78 and 0.75 for RF.\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\u003eMetrics for the classification by classes (cases or controls) for the four supervised algorithms\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"9\"\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=\"left\" 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=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMetric\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u003cp\u003ePrecision\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eRecall\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c7\" namest=\"c6\"\u003e\u003cp\u003eF1-score\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e\u003cp\u003eSupport\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colspan=\"9\" nameend=\"c9\" namest=\"c1\"\u003e\u003cp\u003eclass\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\u003eModel\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.95\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.86\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.92\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.90\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.85\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e21\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e12\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLR\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.21\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.58\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.43\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.29\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e26\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.76\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.75\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.90\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.50\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.83\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.60\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e21\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e12\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.82\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.73\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.86\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.67\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.84\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.70\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e\u003cp\u003e21\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e\u003cp\u003e12\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\u003eAccording to the results, the lack of apparent separation between the groups indicates the need to explore complementary approaches to improve effective discrimination between the two groups. Therefore, supervised learning methods were implemented in this prospective study (Supplementary Information, Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e). These learning methods are based on dividing the dataset into training and test data, for which 4 models were used (SVM, LR, RF, and XGBoost). The SVM model performed better in global metrics, achieving an accuracy of 0.88 compared to the other models. In this sense, the LR algorithm showed an inferior performance in the differentiation of cases and controls despite having an excellent overall metric.\u003c/p\u003e\u003cp\u003eOn the other hand, SVM presented a better accuracy than the different models, which reflects the percentage of correct predictions over the total predictions, clarifying that it is within the test percentage. Furthermore, sensitivity represents the ability of the model to correctly identify cases (positive) and controls (negative). In this case, the best results by class are divided into two models: SVM with a value of 0.92 for cases (1) and RF with a value of 0.90 for controls (0). Finally, the F1-score combines the two previous metrics to choose the best model, which in this case was SVM, which obtained the best performance in differentiating women with and without preeclampsia in a test dataset of 20 percent. However, it is crucial to consider detailed evaluation metrics per class due to their imbalance (97 controls, 67 cases), as shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, when calculating the complete values for all metrics. Although, despite having a good performance in this global metric, limitations in the model may be present, so one should not rely on a single metric to measure the performance of the predictive model, so a thorough analysis of other more detailed classification metrics by class may provide a better understanding of the model performance, these were accuracy, recall, and F1-Score.\u003c/p\u003e\u003cp\u003eOn the other hand, the confusion matrix allows an accurate visual representation of each classification model for the test data. Identifying true and false cases, as well as true and false controls. In the test set, specifically, the 20 percent reserved, it was evident that the RL was the model with the highest error rate during training, as is shown in Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e in the supplementary information. There is a general tendency for the evaluation models to perform better at predicting cases than controls.\u003c/p\u003e\u003cp\u003eHowever, it is essential to highlight that an inferior performance was observed in predicting false cases and controls. These evaluation metrics enabled the establishment of the most effective classification model to distinguish between patients with and without preeclampsia, using blood serum samples that had been previously digested and characterized by MALDI-TOF-MS. The SVM algorithm achieved an accuracy of 0.88, results that are consistent with previous research. For instance, a review of the most widely implemented models for preeclampsia diagnosis emphasized the importance of recall as an evaluation metric in clinical studies, especially when faced with minimal differences between cases and controls [34,35]. Other research reported a high efficiency of RF, followed by SVM and RL models in analyzing clinical data from women with preeclampsia. The SVM model achieved an accuracy of 0.68 and a precision of 0.51, while the RF model had a precision of 0.86 [36].\u003c/p\u003e\u003cp\u003e Compared to other research using intact protein samples, where outstanding results were obtained for binary classification, the high versatility of the data from mass spectrometry in developing predictive mathematical models is evident, which supports the suitability of the methodology used in this study for digested protein samples [37]. In this study, differences in the intensity of the mass spectra between the case and control groups are observed, enabling the identification of distinct patterns between the sample groups of preeclampsia. The SVM model demonstrated superior performance in the cohort of pregnant women in the study. This result contrasts with previous findings where the differences between cases and controls for preeclampsia were minimal, and the SVM performed poorly in predicting [38]. This discrepancy may be due to the complexity of the disease, variations in data analysis methods, and the number of datasets.\u003c/p\u003e\u003cp\u003e\u003cb\u003eLimitations and perspectives.\u003c/b\u003e This study presents a novel approach to the application of machine learning models for the clinical classification of preeclampsia cases. Despite this approach's inherent limitations, such as the need for large data sets, these methods can complement traditional techniques by providing deeper insight into the underlying patterns in mass spectra. This modelling with a big data set improved predictive ability and, ultimately, improved clinical care for pregnant women at risk for preeclampsia. Furthermore, it encourages research in the clinical field by relying on such computational tools and spectrometric data.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThrough this study, we implemented a preparatory method based on the FASP protocol to obtain the peptide fingerprint of blood serum samples from preeclampsia cases and controls, thereby generating high-quality mass spectra. We then formulated a predictive model based on these data using machine learning tools. The algorithm based on support vector machines was the one that obtained the highest accuracy (0.88) and recall (0.92) to predict the true cases as evident in the results of the metrics expressed in the confusion matrix. These results represent a significant advance in disease detection and pave the way for future studies in the timely diagnosis of conditions such as Preeclampsia, supported by machine learning data analysis. In addition, this research demonstrates the versatility of the FASP method in protein sample preparation and data analysis using supervised classification algorithms, which can be applied to the study of other complex multifactorial diseases.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate.\u0026nbsp;\u003c/strong\u003eThis study was conducted using samples collected in the GenPE project, which established a biobank of information and human biological material (maternal and neonatal) through the active recruitment of cases and controls over a 12-year period (2000-2012). During this recruitment process and before entering the study, written informed consent was obtained from the pregnant women to participate in the research and to store a blood sample and its derivatives for analysis related to the study of preeclampsia.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability.\u0026nbsp;\u003c/strong\u003eLabels for case and control samples are summarized in Table S1 on the Information Support used in this study, with the MALDI spectral data reservoir are publicly available on: https://github.com/YAPrada/MALDI-Data-Preeclampsia.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAnd we present in the Information Supporting Figure S1. Confusion matrices for all machine learning algorithms for the classification of Preeclampsia cases.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode Availability.\u0026nbsp;\u003c/strong\u003eThe codes are available to public access as a notebook on: https://github.com/Lauracmaca/MODELO-PE-MALDI-TOF/blob/main/Support%20Information%20S1%205-7-2024%20.ipynb.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests.\u0026nbsp;\u003c/strong\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding.\u003c/strong\u003e This work was supported by the OEI Program \u0026lsquo;Women, Science and Equity\u0026rsquo; through the Ministry of Science, Technology and Innovation (MinCiencias), Colombia.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; Contributions.\u0026nbsp;\u003c/strong\u003eLA\u003csup\u003e\u0026dagger;\u003c/sup\u003e and LM\u003csup\u003e\u0026dagger;\u003c/sup\u003e performed the experimental probes, training/testing datasets, and wrote the manuscript. YAP, PB, and CC supervised the investigation, data interpretation, and manuscript review and editing. NC, DC, and EMO designed this study and reviewed the manuscript. All authors have approved the final version of the manuscript. \u003csup\u003e\u0026dagger;\u003c/sup\u003eThese authors contributed equally\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e. We thank the Mass Spectrometry Laboratory of Parque Tecnol\u0026oacute;gico Guatiguar\u0026aacute; at Universidad Industrial de Santander for allowing us to use the mass spectrometer for the spectral analysis.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSammour MB, El-Kabarity H, Fawzy MM, Schindler a. E. Prevention and treatment of pre-eclampsia and eclampsia. Journal of Steroid Biochemistry \u0026amp; Molecular Biology. 2011. \u003c/li\u003e\n\u003cli\u003eNavajas R, Corrales F, Paradela A. Quantitative proteomics-based analyses performed on pre-eclampsia samples in the 2004\u0026ndash;2020 period: a systematic review. Clin Proteomics [Internet]. 2021;18:1\u0026ndash;12. Available from: https://doi.org/10.1186/s12014-021-09313-1\u003c/li\u003e\n\u003cli\u003eLu HQ, Hu R. Lasting effects of intrauterine exposure to preeclampsia on offspring and the underlying mechanism. AJP Rep. 2019;9:E275\u0026ndash;91. \u003c/li\u003e\n\u003cli\u003eBellamy L, Casas JP, Hingorani AD, Williams DJ. Pre-eclampsia and risk of cardiovascular disease and cancer in later life: Systematic review and meta-analysis. Br Med J. 2007;335:974\u0026ndash;7. \u003c/li\u003e\n\u003cli\u003eAtluri N, Beyuo TK, Oppong SA, Moyer CA, Lawrence ER. Challenges to diagnosing and managing preeclampsia in a low-resource setting: A qualitative study of obstetric provider perspectives from Ghana. PLOS Global Public Health [Internet]. 2023;3:e0001790. Available from: http://dx.doi.org/10.1371/journal.pgph.0001790\u003c/li\u003e\n\u003cli\u003eOpichka MA, Rappelt MW, Gutterman DD, Grobe JL, McIntosh JJ. Review vascular dysfunction in preeclampsia. Cells. 2021;10. \u003c/li\u003e\n\u003cli\u003eSahai K, Saraswathy S, Yadav TP, Arora D, Krishnan M. Pre-eclampsia: Molecular events to biomarkers. Med J Armed Forces India [Internet]. 2017;73:167\u0026ndash;74. Available from: http://dx.doi.org/10.1016/j.mjafi.2016.09.001\u003c/li\u003e\n\u003cli\u003eNobakht M. Gh BF. Application of metabolomics to preeclampsia diagnosis. Syst Biol Reprod Med [Internet]. 2018;64:324\u0026ndash;39. Available from: https://doi.org/10.1080/19396368.2018.1482968\u003c/li\u003e\n\u003cli\u003eTarca AL, Romero R, Benshalom-Tirosh N, Than NG, Gudicha DW, Done B, et al. The prediction of early preeclampsia: Results from a longitudinal proteomics study. PLoS One. 2019. \u003c/li\u003e\n\u003cli\u003eGhaemi MS, Tarca AL, Romero R, Stanley N, Fallahzadeh R, Tanada A, et al. Proteomic signatures predict preeclampsia in individual cohorts but not across cohorts\u0026ndash;implications for clinical biomarker studies. Journal of Maternal-Fetal and Neonatal Medicine [Internet]. 2022;35:5621\u0026ndash;8. Available from: https://doi.org/10.1080/14767058.2021.1888915\u003c/li\u003e\n\u003cli\u003eThan NG, Romero R, Gy\u0026ouml;rffy D, Posta M, Bhatti G, Done B, et al. Molecular subclasses of preeclampsia characterized by a longitudinal maternal proteomics study: Distinct biomarkers, disease pathways and options for prevention. J Perinat Med. 2023;51:51\u0026ndash;68. \u003c/li\u003e\n\u003cli\u003eBanach P, Dereziński P, Matuszewska E, Matysiak J, Bochyński H, Kokot ZJ, et al. MALDI-TOF-MS analysis in the identification of urine proteomic patterns of gestational trophoblastic disease. Metabolites. 2019;9. \u003c/li\u003e\n\u003cli\u003eRozanova S, Barkovits K, Nikolov M, Schmidt C, Urlaub H, Marcus K. Quantitative Mass Spectrometry-Based Proteomics: An Overview. Methods in Molecular Biology. 2021. \u003c/li\u003e\n\u003cli\u003eZaikin VG, Borisov RS. Mass Spectrometry as a Crucial Analytical Basis for Omics Sciences. Journal of Analytical Chemistry. 2021;76:1567\u0026ndash;87. \u003c/li\u003e\n\u003cli\u003eDarie-Ion L, Whitham D, Jayathirtha M, Rai Y, Neagu AN, Darie CC, et al. Applications of MALDI-MS/MS-Based Proteomics in Biomedical Research. Molecules. 2022;27:1\u0026ndash;23. \u003c/li\u003e\n\u003cli\u003eKlein J, Buffin-Meyer B, Mullen W, Carty DM, Delles C, Vlahou A, et al. Clinical proteomics in obstetrics and neonatology. Expert Rev Proteomics. 2014;11:75\u0026ndash;89. \u003c/li\u003e\n\u003cli\u003eChen H, Aneman I, Nikolic V, Karadzov Orlic N, Mikovic Z, Stefanovic M, et al. Maternal plasma proteome profiling of biomarkers and pathogenic mechanisms of early-onset and late-onset preeclampsia. Sci Rep [Internet]. 2022;12:1\u0026ndash;13. Available from: https://doi.org/10.1038/s41598-022-20658-x\u003c/li\u003e\n\u003cli\u003eBujold E, Fillion A, Roux-Dalvai F, Scott-Boyer MP, Gigu\u0026egrave;re Y, Forest JC, et al. Proteomic analysis of maternal urine for the early detection of preeclampsia and fetal growth restriction. J Clin Med. 2021;10. \u003c/li\u003e\n\u003cli\u003eDavodabadi A, Daneshian B, Saati S, Razavyan S. Mathematical model and artificial intelligence for diagnosis of Alzheimer\u0026rsquo;s disease. Eur Phys J Plus [Internet]. 2023;138. Available from: https://doi.org/10.1140/epjp/s13360-023-04128-5\u003c/li\u003e\n\u003cli\u003eSedaghati F, Gleason RL. A mathematical model of vascular and hemodynamics changes in early and late forms of preeclampsia. Physiol Rep. 2023;11:1\u0026ndash;21. \u003c/li\u003e\n\u003cli\u003eM\u0026eacute;sz\u0026aacute;ros B, Kukor Z, Valent S. Recent Advances in the Prevention and Screening of Preeclampsia. J Clin Med. 2023;12. \u003c/li\u003e\n\u003cli\u003eCarreno JF, Qiu P. Feature selection algorithms for predicting preeclampsia: A comparative approach. Proceedings - 2020 IEEE International Conference on Bioinformatics and Biomedicine, BIBM 2020. 2020;2626\u0026ndash;31. \u003c/li\u003e\n\u003cli\u003eLiu M, Yang X, Chen G, Ding Y, Shi M, Sun L, et al. Development of a prediction model on preeclampsia using machine learning-based method: a retrospective cohort study in China. Front Physiol. 2022;13:1\u0026ndash;9. \u003c/li\u003e\n\u003cli\u003eSarkar D, Saha S. Machine-learning techniques for the prediction of protein\u0026ndash;protein interactions. J Biosci. 2019;44. \u003c/li\u003e\n\u003cli\u003eAljameel SS, Alzahrani M, Almusharraf R, Altukhais M, Alshaia S, Sahlouli H, et al. Prediction of Preeclampsia Using Machine Learning and Deep Learning Models: A Review. Big Data and Cognitive Computing. 2023;7. \u003c/li\u003e\n\u003cli\u003eThompson J, Everhart Nunn SL, Sarkar S, Clayton B. Diagnostic Screening of Bovine Mastitis Using MALDI-TOF MS Direct-Spotting of Milk and Machine Learning. Vet Sci. 2023;10. \u003c/li\u003e\n\u003cli\u003eSerrano NC, Guio-Mahecha E, Quintero-Lesmes DC, Becerra-Bayona S, Paez MC, Beltran M, et al. Lipid profile, plasma apolipoproteins, and pre-eclampsia risk in the GenPE case-control study. Atherosclerosis. 2018;276:189\u0026ndash;94. \u003c/li\u003e\n\u003cli\u003eColmenares-Mejia CC, Quintero-Lesmes DC, Bautista-Ni\u0026ntilde;o PK, Gu\u0026iacute;o E, Paez MC, Beltr\u0026aacute;n M, et al. Uric acid and risk of pre-eclampsia: results from a large case\u0026ndash;control study and meta-analysis of prospective studies. Sci Rep [Internet]. 2023;13:1\u0026ndash;10. Available from: https://doi.org/10.1038/s41598-023-29651-4\u003c/li\u003e\n\u003cli\u003eWiśniewski JR. Filter-aided sample preparation for proteome analysis. Methods in Molecular Biology. 2018;1841:3\u0026ndash;10. \u003c/li\u003e\n\u003cli\u003eDaltonics B. Bruker Guide to MALDI Sample Preparation [Internet]. MALDI Preparation Protocols. 2012. Available from: www.care-bdal.com\u003c/li\u003e\n\u003cli\u003eDave K a, Headlam MJ, Wallis TP, Gorman JJ. Preparation and analysis of proteins and peptides using MALDI TOF/TOF mass spectrometry. Current protocols in protein science / editorial board, John E Coligan . [et al]. 2011;Chapter 16:Unit 16.13. \u003c/li\u003e\n\u003cli\u003eLiu Y, Wu R, Yang A. Research on Medical Problems Based on Mathematical Models. Mathematics. 2023;11. \u003c/li\u003e\n\u003cli\u003eKe M, Shen H, Wang L, Luo S, Lin L, Yang J, et al. Sample Preparation, Analysis and Practical Applications for mass spectrometry-based proteomics [Internet]. Adv Exp Med Biol. 2016. Available from: http://www.springer.com/gp/book/9783319414461%0Ahttp://link.springer.com/10.1007/978-3-319-41448-5\u003c/li\u003e\n\u003cli\u003eSchauberger G, Klug SJ, Berger M. Random forests for the analysis of matched case\u0026ndash;control studies. BMC Bioinformatics. 2024;25. \u003c/li\u003e\n\u003cli\u003eHuang S, Nianguang CAI, Penzuti Pacheco P, Narandes S, Wang Y, Wayne XU. Applications of support vector machine (SVM) learning in cancer genomics. Cancer Genomics Proteomics. International Institute of Anticancer Research; 2018. p. 41\u0026ndash;51. \u003c/li\u003e\n\u003cli\u003eZhang D, Hu Y, Guo W, Song Y, Yang L, Yang S, et al. Mendelian randomization study reveals a causal relationship between rheumatoid arthritis and risk for pre-eclampsia. Front Immunol. 2022;13:1\u0026ndash;8. \u003c/li\u003e\n\u003cli\u003eYang J, Gao Z, Ren X, Sheng J, Xu P, Chang C, et al. DeepDigest: Prediction of Protein Proteolytic Digestion with Deep Learning. Anal Chem. 2021;93:6094\u0026ndash;103. \u003c/li\u003e\n\u003cli\u003eJhee JH, Lee S, Park Y, Lee SE, Kim YA, Kang SW, et al. Prediction model development of late-onset preeclampsia using machine learning-based methods. PLoS One. 2019;14:1\u0026ndash;12. \u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Preeclampsia, Protein-fingerprinting, Proteomics, Mass spectrometry, Machine learning","lastPublishedDoi":"10.21203/rs.3.rs-6966953/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6966953/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003ePreeclampsia is a pregnancy-induced disorder characterized by hypertension and high levels of proteinuria after 20 weeks of pregnancy. This condition significantly raises the risk of maternal-fetal death and increases the risk of vascular diseases after pregnancy. However, timely and unequivocal diagnosis in the first weeks allows access to appropriate medical follow-up. In this work, we performed a supervised classificatory analysis by machine learning using the protein profiles present in blood serum samples from 97 controls and 67 patient cases obtained by matrix-assisted laser ionization/desorption time-of-flight mass spectrometry.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eThe protein profile was obtained from peptide mass fingerprinting of the samples by the Filter-Assisted Sample Preparation (FASP) protocol and subsequent acquisition of the mass spectra. The spectrum data analysis using machine learning algorithms demonstrated high performance in classifying cases and controls, with an overall accuracy of 88% and a sensitivity of 0.90 and 0.85 for predicting positive cases and negative controls, respectively.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e\u003cp\u003eThese results highlight the reliability and versatility of analyzing and processing spectral data from mass spectrometry using artificial intelligence tools to study of preeclampsia. This study could pave the way for future applications in clinical diagnostics, offering new alternatives for improved patient outcomes.\u003c/p\u003e","manuscriptTitle":"Supervised classification of Preeclampsia clinical cases using datasets from MALDI-TOF-MS and machine learning tools","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-14 10:15:36","doi":"10.21203/rs.3.rs-6966953/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4fb6b1ce-aa02-43d4-ae72-6d25ea7a83b5","owner":[],"postedDate":"July 14th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-11-22T19:38:15+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-14 10:15:36","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6966953","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6966953","identity":"rs-6966953","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}