{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"medRxiv"}],"indexed":{"date-parts":[[2026,2,26]],"date-time":"2026-02-26T18:06:13Z","timestamp":1772129173235,"version":"3.50.1"},"posted":{"date-parts":[[2026,2,24]]},"group-title":"Ophthalmology","reference-count":65,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2026,2,24]],"date-time":"2026-02-24T00:00:00Z","timestamp":1771891200000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2026,2,24]]},"abstract":"<jats:title>ABSTRACT<\/jats:title>\n                <jats:sec>\n                  <jats:title>Purpose<\/jats:title>\n                  <jats:p>There is a need for novel therapies for diabetic retinopathy (DR) because existing therapies treat only certain features of DR and do not work optimally for all patients. While proteomic studies provide insight into disease pathobiology, they are often limited to small sample sizes due to high costs, limiting their generalizability and reproducibility. Moreover, they often yield lists of tens to hundreds of proteins with differential expression, making it difficult to prioritize the most biologically relevant biomarkers beyond using arbitrary fold-change and false-detection rate cutoffs. Here, we applied a two-stage multimodal AI approach: first, we integrated EHR and proteomics data to rationally prioritize candidate protein biomarkers and, next, validated these biomarkers in an independent cohort. These protein biomarkers of DR are rooted in the EHR data and thereby more likely to be biological drivers of disease.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Methods<\/jats:title>\n                  <jats:p>\n                    We obtained EHR data from a large number of patients with and without DR (N=319,997) from the STARR-OMOP database and obtained aqueous humor liquid biopsies from a subset of these patients (N=101) for high-resolution proteomic profiling. We developed\n                    <jats:bold>C<\/jats:bold>\n                    linical and\n                    <jats:bold>O<\/jats:bold>\n                    mics\n                    <jats:bold>M<\/jats:bold>\n                    ulti-Modal Analysis\n                    <jats:bold>E<\/jats:bold>\n                    nhanced with\n                    <jats:bold>T<\/jats:bold>\n                    ransfer Learning (COMET) to perform integrated analysis of proteomics and all available EHR data to identify protein biomarkers of DR. The model was trained in two phases: first, it was pretrained using patients with EHR data alone (N=319,896), and then, it was fine tuned using patients with both EHR and proteomics data (N=101), allowing it to learn both clinical and molecular features associated with DR. Findings from COMET were then validated with liquid biopsies from an independent, validation cohort (N=164).\n                  <\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Results<\/jats:title>\n                  <jats:p>t-distributed stochastic neighbor embedding (t-SNE) analysis of EHR and proteomics data identified proteins clustering with related EHR features. Levels of STX3 and NOTCH2, proteins involved in retinal function, were correlated with a diagnosis of macular edema, a record of a visual field exam, and a prescription for latanoprost, highlighting protein-EHR alignment. The pretrained, multimodal COMET model was superior (AUROC=0.98, AUPRC=0.91) compared to models generated using either EHR or proteomics data alone or without pretraining (AUROC: 0.76 to 0.92; AUPRC: 0.47 to 0.74). The proteins SERPINE1, QPCT, AKR1C2, IL2RB, and SRSF6 were prioritized by the COMET model compared to the models without pretraining, supporting their potential role in DR pathobiology, and were subsequently validated in an independent cohort.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Conclusion<\/jats:title>\n                  <jats:p>We used multimodal AI to prioritize protein biomarkers of DR that are most strongly linked to EHR elements, as well as identifying other protein biomarkers associated with disease features like diabetic macular edema. These findings serve as a foundation for future mechanistic studies and highlight the synergistic value of using multimodal AI to fuse EHR and proteomics data for enhanced proteomics analysis.<\/jats:p>\n                <\/jats:sec>","DOI":"10.64898\/2026.02.23.26346903","type":"posted-content","created":{"date-parts":[[2026,2,25]],"date-time":"2026-02-25T03:00:20Z","timestamp":1771988420000},"source":"Crossref","is-referenced-by-count":0,"title":["Multimodal AI fuses proteomic and EHR data for rational prioritization of protein biomarkers in diabetic retinopathy"],"prefix":"10.64898","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-2792-7821","authenticated-orcid":false,"given":"Jonathan B.","family":"Lin","sequence":"first","affiliation":[{"name":"Molecular Surgery Laboratory, Department of Ophthalmology, Stanford University, Palo Alto, CA 94304, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Samson J.","family":"Mataraso","sequence":"additional","affiliation":[{"name":"Department of Anesthesiology, Perioperative and Pain Medicine, Stanford University School of Medicine, Stanford, CA 94305, USA"},{"name":"Department of Pediatrics, Stanford University School of Medicine, Stanford, CA 94305, USA"},{"name":"Department of Biomedical Data Science, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Madhumeeta","family":"Chadha","sequence":"additional","affiliation":[{"name":"Molecular Surgery Laboratory, Department of Ophthalmology, Stanford University, Palo Alto, CA 94304, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Gabriel","family":"Velez","sequence":"additional","affiliation":[{"name":"Molecular Surgery Laboratory, Department of Ophthalmology, Stanford University, Palo Alto, CA 94304, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Prithvi","family":"Mruthyunjaya","sequence":"additional","affiliation":[{"name":"Molecular Surgery Laboratory, Department of Ophthalmology, Stanford University, Palo Alto, CA 94304, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Nima","family":"Aghaeepour","sequence":"additional","affiliation":[{"name":"Department of Anesthesiology, Perioperative and Pain Medicine, Stanford University School of Medicine, Stanford, CA 94305, USA"},{"name":"Department of Pediatrics, Stanford University School of Medicine, Stanford, CA 94305, USA"},{"name":"Department of Biomedical Data Science, Stanford University School of Medicine, Stanford, CA 94305, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Vinit B.","family":"Mahajan","sequence":"additional","affiliation":[{"name":"Molecular Surgery Laboratory, Department of Ophthalmology, Stanford University, Palo Alto, CA 94304, USA"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"54368","reference":[{"key":"2026022609153728000_2026.02.23.26346903v1.1","doi-asserted-by":"crossref","unstructured":"Duh, E. J. , Sun, J. K. & Stitt, A. W . Diabetic retinopathy: current understanding, mechanisms, and treatment strategies. JCI Insight 2, (2017).","DOI":"10.1172\/jci.insight.93751"},{"key":"2026022609153728000_2026.02.23.26346903v1.2","doi-asserted-by":"crossref","first-page":"2042018821997320","DOI":"10.1177\/2042018821997320","article-title":"GLP-1 receptor agonists: an updated review of head-to-head clinical studies","volume":"12","year":"2021","journal-title":"Ther. Adv. Endocrinol. Metab"},{"key":"2026022609153728000_2026.02.23.26346903v1.3","doi-asserted-by":"publisher","DOI":"10.1056\/NEJM199309303291401"},{"key":"2026022609153728000_2026.02.23.26346903v1.4","doi-asserted-by":"publisher","DOI":"10.1016\/S0140-6736(98)07019-6"},{"key":"2026022609153728000_2026.02.23.26346903v1.5","doi-asserted-by":"crossref","first-page":"747","DOI":"10.1001\/jamaophthalmol.2023.2289","article-title":"Prevalence of diabetic retinopathy in the US in 2021","volume":"141","year":"2023","journal-title":"JAMA Ophthalmol"},{"key":"2026022609153728000_2026.02.23.26346903v1.6","doi-asserted-by":"crossref","first-page":"599","DOI":"10.1001\/jamaophthalmol.2024.0046","article-title":"Longitudinal trends and disparities in diabetic retinopathy within an aggregate health care network","volume":"142","year":"2024","journal-title":"JAMA Ophthalmol"},{"key":"2026022609153728000_2026.02.23.26346903v1.7","doi-asserted-by":"crossref","first-page":"2778","DOI":"10.3390\/jcm13102778","article-title":"Beyond VEGF: Angiopoietin-Tie signaling pathway in diabetic retinopathy","volume":"13","year":"2024","journal-title":"J. Clin. Med"},{"key":"2026022609153728000_2026.02.23.26346903v1.8","doi-asserted-by":"publisher","DOI":"10.1124\/jpet.122.001563"},{"key":"2026022609153728000_2026.02.23.26346903v1.9","doi-asserted-by":"publisher","DOI":"10.1038\/s41598-018-26640-w"},{"key":"2026022609153728000_2026.02.23.26346903v1.10","doi-asserted-by":"publisher","DOI":"10.1038\/s10038-020-0763-5"},{"key":"2026022609153728000_2026.02.23.26346903v1.11","doi-asserted-by":"publisher","DOI":"10.3390\/metabo10070286"},{"key":"2026022609153728000_2026.02.23.26346903v1.12","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pone.0197910"},{"key":"2026022609153728000_2026.02.23.26346903v1.13","first-page":"e893","article-title":"Leveraging electronic health records for data science: common pitfalls and how to avoid them. Lancet Digit","volume":"4","year":"2022","journal-title":"Health"},{"key":"2026022609153728000_2026.02.23.26346903v1.14","doi-asserted-by":"crossref","first-page":"100037","DOI":"10.1016\/j.xops.2021.100037","article-title":"Temporal trends in the treatment of proliferative diabetic retinopathy: An AAO IRIS\u00ae registry analysis","volume":"1","year":"2021","journal-title":"Ophthalmol. Sci"},{"key":"2026022609153728000_2026.02.23.26346903v1.15","doi-asserted-by":"crossref","unstructured":"Huang, Q. et al. Biomarker Identification by Proteomic Analysis of Vitreous Humor and Plasma in Diabetic Retinopathy. Physiology (2024).","DOI":"10.1101\/2024.05.18.594835"},{"key":"2026022609153728000_2026.02.23.26346903v1.16","doi-asserted-by":"crossref","unstructured":"Stahlschmidt, S. R. , Ulfenborg, B. & Synnergren, J . Multimodal deep learning for biomedical data fusion: a review. Brief. Bioinform. 23, (2022).","DOI":"10.1093\/bib\/bbab569"},{"key":"2026022609153728000_2026.02.23.26346903v1.17","doi-asserted-by":"crossref","first-page":"293","DOI":"10.1038\/s42256-024-00974-9","article-title":"A machine learning approach to leveraging electronic health records for enhanced omics analysis. Nat","volume":"7","year":"2025","journal-title":"Mach. Intell"},{"key":"2026022609153728000_2026.02.23.26346903v1.18","doi-asserted-by":"publisher","DOI":"10.1016\/j.cell.2023.09.012"},{"key":"2026022609153728000_2026.02.23.26346903v1.19","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.2010751117"},{"key":"2026022609153728000_2026.02.23.26346903v1.20","doi-asserted-by":"publisher","DOI":"10.1007\/s00439-021-02284-1"},{"key":"2026022609153728000_2026.02.23.26346903v1.21","first-page":"78","article-title":"The expression of NOTCH2, HES1 and SOX9 during mouse retinal development. Gene Expr","volume":"13","year":"2013","journal-title":"Patterns"},{"key":"2026022609153728000_2026.02.23.26346903v1.22","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.1814711116"},{"key":"2026022609153728000_2026.02.23.26346903v1.23","doi-asserted-by":"publisher","DOI":"10.1210\/me.2005-0197"},{"key":"2026022609153728000_2026.02.23.26346903v1.24","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.1703.01365"},{"key":"2026022609153728000_2026.02.23.26346903v1.25","doi-asserted-by":"crossref","first-page":"442","DOI":"10.55730\/1300-0152.2718","article-title":"Characterization of TFIIE-regulated genes by transcriptome analysis","volume":"48","year":"2024","journal-title":"Turk. J. Biol"},{"key":"2026022609153728000_2026.02.23.26346903v1.26","doi-asserted-by":"publisher","DOI":"10.1021\/acs.biochem.4c00489"},{"key":"2026022609153728000_2026.02.23.26346903v1.27","doi-asserted-by":"publisher","DOI":"10.1074\/jbc.M113.545582"},{"key":"2026022609153728000_2026.02.23.26346903v1.28","first-page":"1","article-title":"High expression of CIN85 promotes proliferation and invasion of human esophageal squamous cell carcinoma","volume":"23","year":"2021","journal-title":"Mol. Med. Rep"},{"key":"2026022609153728000_2026.02.23.26346903v1.29","doi-asserted-by":"crossref","first-page":"506","DOI":"10.1007\/s10120-024-01480-y","article-title":"Killer cell lectin-like receptor G2 facilitates aggressive phenotypes of gastric cancer cells via dual activation of the ERK1\/2 and JAK\/STAT pathways","volume":"27","year":"2024","journal-title":"Gastric Cancer"},{"key":"2026022609153728000_2026.02.23.26346903v1.30","doi-asserted-by":"publisher","DOI":"10.1186\/1476-4598-12-127"},{"key":"2026022609153728000_2026.02.23.26346903v1.31","doi-asserted-by":"publisher","DOI":"10.3389\/fphys.2017.00887"},{"key":"2026022609153728000_2026.02.23.26346903v1.32","doi-asserted-by":"publisher","DOI":"10.1002\/jcp.22279"},{"key":"2026022609153728000_2026.02.23.26346903v1.33","doi-asserted-by":"publisher","DOI":"10.1242\/dmm.031864"},{"key":"2026022609153728000_2026.02.23.26346903v1.34","doi-asserted-by":"publisher","DOI":"10.1038\/s41419-019-1455-y"},{"key":"2026022609153728000_2026.02.23.26346903v1.35","doi-asserted-by":"publisher","DOI":"10.3390\/ijms22063170"},{"key":"2026022609153728000_2026.02.23.26346903v1.36","doi-asserted-by":"publisher","DOI":"10.1046\/j.1538-7836.2003.00279.x"},{"key":"2026022609153728000_2026.02.23.26346903v1.37","first-page":"2746","article-title":"Plasminogen activator inhibitor-1 gene-deficient mice","volume":"92","year":"1993","journal-title":"I. Generation by homologous recombination and characterization. J. Clin. Invest"},{"key":"2026022609153728000_2026.02.23.26346903v1.38","doi-asserted-by":"publisher","DOI":"10.1167\/iovs.09-3619"},{"key":"2026022609153728000_2026.02.23.26346903v1.39","doi-asserted-by":"publisher","DOI":"10.1016\/j.diabres.2009.11.007"},{"key":"2026022609153728000_2026.02.23.26346903v1.40","doi-asserted-by":"crossref","first-page":"eabm1896","DOI":"10.1126\/sciadv.abm1896","article-title":"PAI-1 is a vascular cell-specific HIF-2-dependent angiogenic factor that promotes retinal neovascularization in diabetic patients","volume":"8","year":"2022","journal-title":"Sci. Adv"},{"key":"2026022609153728000_2026.02.23.26346903v1.41","doi-asserted-by":"publisher","DOI":"10.1016\/j.drudis.2023.103644"},{"key":"2026022609153728000_2026.02.23.26346903v1.42","doi-asserted-by":"crossref","first-page":"1209863","DOI":"10.3389\/fnagi.2023.1209863","article-title":"Development and evolution of human glutaminyl cyclase inhibitors (QCIs): an alternative promising approach for disease-modifying treatment of Alzheimer\u2019s disease","volume":"15","year":"2023","journal-title":"Front. Aging Neurosci"},{"key":"2026022609153728000_2026.02.23.26346903v1.43","doi-asserted-by":"crossref","first-page":"S79","DOI":"10.3233\/JAD-231126","article-title":"Varoglutamstat: Inhibiting glutaminyl cyclase as a novel target of therapy in early Alzheimer\u2019s disease","volume":"101","year":"2024","journal-title":"J. Alzheimers. Dis"},{"key":"2026022609153728000_2026.02.23.26346903v1.44","doi-asserted-by":"crossref","first-page":"11791","DOI":"10.3390\/ijms222111791","article-title":"Combination of the glutaminyl cyclase inhibitor PQ912 (Varoglutamstat) and the Murine monoclonal antibody PBD-C06 (m6) shows additive effects on brain A\u03b2 pathology in transgenic mice","volume":"22","year":"2021","journal-title":"Int. J. Mol. Sci."},{"key":"2026022609153728000_2026.02.23.26346903v1.45","doi-asserted-by":"crossref","first-page":"109513","DOI":"10.1016\/j.exer.2023.109513","article-title":"Identification and comprehensive analysis of ferroptosis-related genes as potential biomarkers for the diagnosis and treatment of proliferative diabetic retinopathy by bioinformatics methods","volume":"232","year":"2023","journal-title":"Exp. Eye Res"},{"key":"2026022609153728000_2026.02.23.26346903v1.46","doi-asserted-by":"publisher","DOI":"10.1016\/S0969-9961(03)00101-3"},{"key":"2026022609153728000_2026.02.23.26346903v1.47","doi-asserted-by":"publisher","DOI":"10.1016\/j.exer.2005.10.021"},{"key":"2026022609153728000_2026.02.23.26346903v1.48","doi-asserted-by":"publisher","DOI":"10.1167\/iovs.63.10.3"},{"key":"2026022609153728000_2026.02.23.26346903v1.49","first-page":"985","article-title":"IL-2 and IFN-gamma in the retina of diabetic rats","volume":"248","year":"2010","journal-title":"Arbeitsphysiologie"},{"key":"2026022609153728000_2026.02.23.26346903v1.50","doi-asserted-by":"crossref","first-page":"109194","DOI":"10.1016\/j.exer.2022.109194","article-title":"Integrative study of gene expression datasets in retinal samples of Diabetic Retinopathy","volume":"223","year":"2022","journal-title":"Exp. Eye Res"},{"key":"2026022609153728000_2026.02.23.26346903v1.51","doi-asserted-by":"publisher","DOI":"10.1016\/0165-0173(95)00015-1"},{"key":"2026022609153728000_2026.02.23.26346903v1.52","doi-asserted-by":"crossref","first-page":"1701","DOI":"10.1007\/s11064-021-03313-1","article-title":"Increased retinal ganglion cell survival by exogenous IL-2 depends on IL-10, dopamine D1 receptors, and classical IL-2\/IL-2R signaling pathways","volume":"46","year":"2021","journal-title":"Neurochem. Res"},{"key":"2026022609153728000_2026.02.23.26346903v1.53","doi-asserted-by":"publisher","DOI":"10.1016\/j.jneuroim.2010.12.008"},{"key":"2026022609153728000_2026.02.23.26346903v1.54","first-page":"1851","article-title":"SRSF6 regulates alternative splicing of genes involved in DNA damage response and DNA repair in HeLa cells","volume":"44","year":"2020","journal-title":"Oncol. Rep"},{"key":"2026022609153728000_2026.02.23.26346903v1.55","doi-asserted-by":"crossref","first-page":"eado8231","DOI":"10.1126\/sciadv.ado8231","article-title":"SRSF6 modulates histone-chaperone HIRA splicing to orchestrate AR and E2F activity in prostate cancer","volume":"10","year":"2024","journal-title":"Sci. Adv"},{"key":"2026022609153728000_2026.02.23.26346903v1.56","doi-asserted-by":"crossref","first-page":"257","DOI":"10.1186\/s12935-024-03438-7","article-title":"Function of serine\/arginine-rich splicing factors in hematopoiesis and hematopoietic malignancies","volume":"24","year":"2024","journal-title":"Cancer Cell Int"},{"key":"2026022609153728000_2026.02.23.26346903v1.57","doi-asserted-by":"publisher","DOI":"10.3389\/fphys.2018.01310"},{"key":"2026022609153728000_2026.02.23.26346903v1.58","doi-asserted-by":"crossref","unstructured":"Stevens, M. & Oltean, S . Modulation of VEGF-A alternative splicing as a novel treatment in chronic kidney disease. Genes (Basel) 9, (2018).","DOI":"10.3390\/genes9020098"},{"key":"2026022609153728000_2026.02.23.26346903v1.59","doi-asserted-by":"publisher","DOI":"10.1046\/j.1523-1755.2000.07718.x"},{"key":"2026022609153728000_2026.02.23.26346903v1.60","doi-asserted-by":"publisher","DOI":"10.1016\/j.jclinepi.2010.04.002"},{"key":"2026022609153728000_2026.02.23.26346903v1.61","doi-asserted-by":"crossref","first-page":"1317","DOI":"10.1038\/s41467-024-55635-7","article-title":"Multi-omic spatial effects on high-resolution AI-derived retinal thickness","volume":"16","year":"2025","journal-title":"Nat. Commun"},{"key":"2026022609153728000_2026.02.23.26346903v1.62","doi-asserted-by":"crossref","first-page":"e0316112","DOI":"10.1371\/journal.pone.0316112","article-title":"The severity of coronary heart disease and the incidence of later diabetic retinopathy in diabetic population: A retrospective cohort study","volume":"20","year":"2025","journal-title":"PLoS One"},{"key":"2026022609153728000_2026.02.23.26346903v1.63","doi-asserted-by":"publisher","DOI":"10.1093\/oxfordjournals.aje.a009786"},{"key":"2026022609153728000_2026.02.23.26346903v1.64","doi-asserted-by":"crossref","first-page":"153","DOI":"10.1186\/s12874-023-01973-x","article-title":"Increase in power by obtaining 10 or more controls per case when type-1 error is small in large-scale association studies","volume":"23","year":"2023","journal-title":"BMC Med. Res. Methodol"},{"key":"2026022609153728000_2026.02.23.26346903v1.65","unstructured":"Datta, S. , et al. A new paradigm for accelerating clinical data science at Stanford Medicine. arXiv [cs.CY] (2020)."}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2026.02.23.26346903","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,2,26]],"date-time":"2026-02-26T17:15:49Z","timestamp":1772126149000},"score":1,"resource":{"primary":{"URL":"http:\/\/medrxiv.org\/lookup\/doi\/10.64898\/2026.02.23.26346903"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,2,24]]},"references-count":65,"URL":"https:\/\/doi.org\/10.64898\/2026.02.23.26346903","relation":{},"subject":[],"published":{"date-parts":[[2026,2,24]]},"subtype":"preprint"}}