{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"bioRxiv"}],"indexed":{"date-parts":[[2026,9,22]],"date-time":"2026-09-22T19:35:17Z","timestamp":1790105717364,"version":"4.0.1"},"posted":{"date-parts":[[2026,4,19]]},"group-title":"Biophysics","reference-count":62,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2026,4,19]],"date-time":"2026-04-19T00:00:00Z","timestamp":1776556800000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by-nd\/4.0\/"}],"funder":[{"award":["DE-SC0025528"],"award-info":[{"award-number":["DE-SC0025528"]}],"id":[{"id":"https:\/\/ror.org\/01bj3aw27","id-type":"ROR","asserted-by":"publisher"}]},{"DOI":"10.13039\/100000002","name":"National Institutes of Health","doi-asserted-by":"crossref","award":["R35 GM151129"],"award-info":[{"award-number":["R35 GM151129"]}],"id":[{"id":"10.13039\/100000002","id-type":"DOI","asserted-by":"crossref"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2026,5,20]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                <jats:p>Predicting protein-ligand binding is a central challenge in computational drug discovery, and while machine learning (ML) and co-folding methods have advanced rapidly, their ability to generalize beyond training or parameterization regimes remains insufficiently understood. DNA-encoded libraries (DELs) enable ultra-large screening of billions of molecules simultaneously, providing a useful testbed for evaluating these approaches at scale. A recent NeurIPS competition revealed that even top performing ML models trained on DEL data failed at generalizing to out-of-distribution (OOD) chemical space. We investigated whether integrating structural modeling could bridge this generalization gap. We systematically assessed state-of-the-art ML, docking, and co-folding methods including Schrodinger Glide, Rosetta GALigandDock, and Boltz-2 with three biologically diverse protein targets screened against libraries containing multiple DEL synthesis formats. While ML excels in-distribution, OOD hit discrimination is dependent on both the target and ligand context, with no single method consistently dominating. These findings demonstrate that benchmark performance alone is insufficient to predict OOD performance, highlighting the need for system-dependent evaluation of binding prediction methods. We provide an open-source package for assessing protein-ligand prediction methods and analyzing high-throughput screening data: DEL-iver.<\/jats:p>\n                <jats:sec>\n                  <jats:title>Abstract Figure<\/jats:title>\n                  <jats:fig id=\"ufig1\" position=\"float\" fig-type=\"figure\" orientation=\"portrait\">\n                    <jats:graphic xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" xlink:href=\"719394v2_ufig1\" position=\"float\" orientation=\"portrait\"\/>\n                  <\/jats:fig>\n                <\/jats:sec>","DOI":"10.64898\/2026.04.18.719394","type":"posted-content","created":{"date-parts":[[2026,4,20]],"date-time":"2026-04-20T04:25:18Z","timestamp":1776659118000},"source":"Crossref","is-referenced-by-count":1,"title":["Assessing the Generalizability of Machine Learning and Physics-Based Methods with DNA-Encoded Libraries"],"prefix":"10.64898","author":[{"given":"Marissa","family":"Dolorfino","sequence":"first","affiliation":[{"name":"Department of Medicinal Chemistry, University of Michigan"},{"name":"Department of Computational Medicine and Bioinformatics, University of Michigan"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Daniel Santos","family":"Perez","sequence":"additional","affiliation":[{"name":"Department of Medicinal Chemistry, University of Michigan"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yao","family":"Fu","sequence":"additional","affiliation":[{"name":"Department of Medicinal Chemistry, University of Michigan"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shu-Hang","family":"Lin","sequence":"additional","affiliation":[{"name":"Department of Medicinal Chemistry, University of Michigan"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sean","family":"McCarty","sequence":"additional","affiliation":[{"name":"Department of Medicinal Chemistry, University of Michigan"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Matthew J.","family":"O\u2019Meara","sequence":"additional","affiliation":[{"name":"Department of Medicinal Chemistry, University of Michigan"},{"name":"Department of Computational Medicine and Bioinformatics, University of Michigan"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-1327-8541","authenticated-orcid":false,"given":"Terra","family":"Sztain","sequence":"additional","affiliation":[{"name":"Department of Medicinal Chemistry, University of Michigan"},{"name":"Department of Biophysics, University of Michigan"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"54368","reference":[{"key":"2026052211051083000_2026.04.18.719394v2.1","doi-asserted-by":"publisher","DOI":"10.1111\/j.1476-5381.2010.01127.x"},{"key":"2026052211051083000_2026.04.18.719394v2.2","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-023-37572-z"},{"key":"2026052211051083000_2026.04.18.719394v2.3","doi-asserted-by":"publisher","DOI":"10.2174\/157340911795677602"},{"key":"2026052211051083000_2026.04.18.719394v2.4","doi-asserted-by":"publisher","DOI":"10.2174\/157340911793743547"},{"key":"2026052211051083000_2026.04.18.719394v2.5","doi-asserted-by":"crossref","first-page":"2831","DOI":"10.1016\/j.csbj.2022.06.004","article-title":"A Brief Review of Protein-Ligand Interaction Prediction","volume":"20","year":"2022","journal-title":"Comput. Struct. Biotechnol. J"},{"key":"2026052211051083000_2026.04.18.719394v2.6","doi-asserted-by":"publisher","DOI":"10.1038\/S41586-021-03819-2"},{"key":"2026052211051083000_2026.04.18.719394v2.7","doi-asserted-by":"publisher","DOI":"10.1038\/S41586-024-07487-W"},{"key":"2026052211051083000_2026.04.18.719394v2.8","doi-asserted-by":"publisher","DOI":"10.1101\/2025.06.14.659707"},{"key":"2026052211051083000_2026.04.18.719394v2.9","unstructured":"Corso, G. ; St\u00e4rk, H. ; Jing, B. ; Barzilay, R. ; Jaakkola, T. DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking. arXiv [q-bio.BM], 2022. http:\/\/arxiv.org\/abs\/2210.01776."},{"issue":"2","key":"2026052211051083000_2026.04.18.719394v2.10","doi-asserted-by":"crossref","first-page":"195","DOI":"10.1038\/s42256-024-00792-z","article-title":"State-Specific Protein\u2013ligand Complex Structure Prediction with a Multiscale Deep Generative Model","volume":"6","year":"2024","journal-title":"Nat. Mach. Intell"},{"issue":"8","key":"2026052211051083000_2026.04.18.719394v2.11","doi-asserted-by":"crossref","first-page":"1315","DOI":"10.1016\/j.jmgm.2008.01.002","article-title":"The Importance of the Domain of Applicability in QSAR Modeling","volume":"26","year":"2008","journal-title":"J. Mol. Graph. Model"},{"key":"2026052211051083000_2026.04.18.719394v2.12","doi-asserted-by":"publisher","DOI":"10.1039\/d5sc05395a"},{"issue":"19","key":"2026052211051083000_2026.04.18.719394v2.13","doi-asserted-by":"crossref","first-page":"4676","DOI":"10.1039\/C9OB00581A","article-title":"Chemical Composition of DNA-Encoded Libraries, Past Present and Future","volume":"17","year":"2019","journal-title":"Org. Biomol. Chem"},{"issue":"4","key":"2026052211051083000_2026.04.18.719394v2.14","doi-asserted-by":"crossref","first-page":"1265","DOI":"10.1021\/acsptsci.1c00118","article-title":"DNA-Encoded Chemical Libraries: A Comprehensive Review with Succesful Stories and Future Challenges","volume":"4","year":"2021","journal-title":"ACS Pharmacol Transl Sci"},{"key":"2026052211051083000_2026.04.18.719394v2.15","doi-asserted-by":"publisher","DOI":"10.1038\/s43586-022-00184-w"},{"key":"2026052211051083000_2026.04.18.719394v2.16","doi-asserted-by":"publisher","DOI":"10.1021\/jacs.0c09213"},{"key":"2026052211051083000_2026.04.18.719394v2.17","doi-asserted-by":"publisher","DOI":"10.1038\/nchembio.211"},{"key":"2026052211051083000_2026.04.18.719394v2.18","doi-asserted-by":"publisher","DOI":"10.1002\/1873-3468.13068"},{"key":"2026052211051083000_2026.04.18.719394v2.19","doi-asserted-by":"publisher","DOI":"10.1016\/j.bmc.2021.116189"},{"issue":"7","key":"2026052211051083000_2026.04.18.719394v2.20","doi-asserted-by":"crossref","first-page":"1166","DOI":"10.1021\/acsmedchemlett.1c00156","article-title":"DNA-Encoded Library Hit Confirmation: Bridging the Gap between on-DNA and off-DNA Chemistry","volume":"12","year":"2021","journal-title":"ACS Med. Chem. Lett"},{"key":"2026052211051083000_2026.04.18.719394v2.21","doi-asserted-by":"publisher","DOI":"10.1021\/acs.jmedchem.0c00452"},{"issue":"20","key":"2026052211051083000_2026.04.18.719394v2.22","doi-asserted-by":"crossref","first-page":"21635","DOI":"10.1021\/acs.jmedchem.5c01972","article-title":"Enabling Open Machine Learning of Deoxyribonucleic Acid-Encoded Library Selections to Accelerate the Discovery of Small Molecule Protein Binders","volume":"68","year":"2025","journal-title":"J. Med. Chem"},{"key":"2026052211051083000_2026.04.18.719394v2.23","doi-asserted-by":"publisher","DOI":"10.1021\/acs.chemrev.4c00284"},{"issue":"2","key":"2026052211051083000_2026.04.18.719394v2.24","doi-asserted-by":"crossref","first-page":"104629","DOI":"10.1016\/j.drudis.2026.104629","article-title":"Toward Generalizable Predictive Models for DNA-Encoded Libraries","volume":"31","year":"2026","journal-title":"Drug Discov. Today"},{"issue":"24","key":"2026052211051083000_2026.04.18.719394v2.25","doi-asserted-by":"crossref","first-page":"10918","DOI":"10.1039\/D5SC00844A","article-title":"Widespread False Negatives in DNA-Encoded Library Data: How Linker Effects Impair Machine Learning-Based Lead Prediction","volume":"16","year":"2025","journal-title":"Chem. Sci"},{"key":"2026052211051083000_2026.04.18.719394v2.26","doi-asserted-by":"publisher","DOI":"10.1038\/s44386-025-00007-4"},{"issue":"16","key":"2026052211051083000_2026.04.18.719394v2.27","doi-asserted-by":"crossref","first-page":"5120","DOI":"10.1021\/acs.jcim.3c00588","article-title":"Building Block-Based Binding Predictions for DNA-Encoded Libraries","volume":"63","year":"2023","journal-title":"J. Chem. Inf. Model"},{"key":"2026052211051083000_2026.04.18.719394v2.28","unstructured":"Blevins, A. ; Halverson, B. J. ; Wilkinson, N. ; Quigley, I. K. BELKA: The big encoded library for chemical assessment. https:\/\/openreview.net\/pdf?id=zwppB4butE (accessed 2024-12-13)."},{"key":"2026052211051083000_2026.04.18.719394v2.29","unstructured":"NeurIPS 2024 - predict new medicines with BELKA. http:\/\/www.kaggle.com\/competitions\/leash-BELKA\/overview\/description (accessed 2024-07-31)."},{"key":"2026052211051083000_2026.04.18.719394v2.30","doi-asserted-by":"publisher","DOI":"10.1021\/acs.jctc.0c01184"},{"key":"2026052211051083000_2026.04.18.719394v2.31","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pone.0020161"},{"key":"2026052211051083000_2026.04.18.719394v2.32","doi-asserted-by":"publisher","DOI":"10.1021\/jm030644s"},{"key":"2026052211051083000_2026.04.18.719394v2.33","doi-asserted-by":"publisher","DOI":"10.1021\/jm051256o"},{"key":"2026052211051083000_2026.04.18.719394v2.34","unstructured":"1st place solution [UPDATED]. https:\/\/www.kaggle.com\/competitions\/leash-BELKA\/writeups\/victor-shlepov-1st-place-solution-updated (accessed 2026-03-19)."},{"key":"2026052211051083000_2026.04.18.719394v2.35","doi-asserted-by":"publisher","DOI":"10.1021\/ci00057a005"},{"key":"2026052211051083000_2026.04.18.719394v2.36","doi-asserted-by":"publisher","DOI":"10.1021\/ci100050t"},{"key":"2026052211051083000_2026.04.18.719394v2.37","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2312.00752"},{"key":"2026052211051083000_2026.04.18.719394v2.38","unstructured":"[5th solution]: Ensemble of CNN1d, Transformer, Mamba models. https:\/\/www.kaggle.com\/competitions\/leash-BELKA\/writeups\/mamba1-one-fold-lb0-432-5th-solution-ensemble-of-c (accessed 2026-03-19)."},{"issue":"12","key":"2026052211051083000_2026.04.18.719394v2.39","doi-asserted-by":"crossref","first-page":"1256","DOI":"10.1038\/s42256-022-00580-7","article-title":"Large-Scale Chemical Language Representations Capture Molecular Structure and Properties","volume":"4","year":"2022","journal-title":"Nat. Mach. Intell"},{"key":"2026052211051083000_2026.04.18.719394v2.40","unstructured":"Liu, Y. ; Ott, M. ; Goyal, N. ; Du, J. ; Joshi, M. ; Chen, D. ; Levy, O. ; Lewis, M. ; Zettlemoyer, L. ; Stoyanov, V. RoBERTa: A Robustly Optimized BERT Pretraining Approach. arXiv [cs.CL], 2019. http:\/\/arxiv.org\/abs\/1907.11692."},{"key":"2026052211051083000_2026.04.18.719394v2.41","unstructured":"8th place private solution - a single SMILES-based transformer model is all we need. And a proper benchmarking. https:\/\/www.kaggle.com\/competitions\/leash-BELKA\/writeups\/vlad-vinogradov-8th-place-private-solution-a-singl (accessed 2026-03-19)."},{"issue":"4","key":"2026052211051083000_2026.04.18.719394v2.42","doi-asserted-by":"crossref","first-page":"1324","DOI":"10.1021\/acs.jcim.8b00825","article-title":"Interpretation of QSAR Models by Coloring Atoms according to Changes in Predicted Activity: How Robust Is It?","volume":"59","year":"2019","journal-title":"J. Chem. Inf. Model"},{"key":"2026052211051083000_2026.04.18.719394v2.43","doi-asserted-by":"publisher","DOI":"10.1021\/ci010132r"},{"key":"2026052211051083000_2026.04.18.719394v2.44","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1802.03426"},{"key":"2026052211051083000_2026.04.18.719394v2.45","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.1808.06640"},{"key":"2026052211051083000_2026.04.18.719394v2.46","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/p18-1032"},{"key":"2026052211051083000_2026.04.18.719394v2.47","first-page":"2786","article-title":"We Need to Talk about Standard Splits","volume":"2019","year":"2019","journal-title":"Proc. Conf. Assoc. Comput. Linguist. Meet"},{"key":"2026052211051083000_2026.04.18.719394v2.48","doi-asserted-by":"publisher","DOI":"10.1101\/2024.09.20.614179"},{"key":"2026052211051083000_2026.04.18.719394v2.49","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pone.0050769"},{"key":"2026052211051083000_2026.04.18.719394v2.50","doi-asserted-by":"publisher","DOI":"10.1016\/j.jmb.2008.11.010"},{"key":"2026052211051083000_2026.04.18.719394v2.51","doi-asserted-by":"publisher","DOI":"10.1021\/jm0608356"},{"key":"2026052211051083000_2026.04.18.719394v2.52","doi-asserted-by":"publisher","DOI":"10.1021\/ci0503255"},{"key":"2026052211051083000_2026.04.18.719394v2.53","doi-asserted-by":"publisher","DOI":"10.1093\/NAR\/GKAF361"},{"issue":"12","key":"2026052211051083000_2026.04.18.719394v2.54","doi-asserted-by":"crossref","first-page":"4661","DOI":"10.1021\/acs.jcim.4c00232","article-title":"Building Block-Centric Approach to DNA-Encoded Library Design","volume":"64","year":"2024","journal-title":"J. Chem. Inf. Model"},{"key":"2026052211051083000_2026.04.18.719394v2.55","doi-asserted-by":"publisher","DOI":"10.5281\/ZENODO.17746401"},{"key":"2026052211051083000_2026.04.18.719394v2.56","doi-asserted-by":"publisher","DOI":"10.1186\/s13321-016-0148-0"},{"key":"2026052211051083000_2026.04.18.719394v2.57","unstructured":"Corso, G. ; Deng, A. ; Fry, B. ; Polizzi, N. ; Barzilay, R. ; Jaakkola, T. Deep Confident Steps to New Pockets: Strategies for Docking Generalization. ArXiv 2024."},{"issue":"9","key":"2026052211051083000_2026.04.18.719394v2.58","doi-asserted-by":"crossref","first-page":"1571","DOI":"10.1021\/bc1001483","article-title":"Drug Discovery with DNA-Encoded Chemical Libraries","volume":"21","year":"2010","journal-title":"Bioconjug. Chem"},{"issue":"8","key":"2026052211051083000_2026.04.18.719394v2.59","doi-asserted-by":"crossref","first-page":"2427","DOI":"10.1016\/j.bmc.2014.03.001","article-title":"Structural Insights into Binding of Inhibitors to Soluble Epoxide Hydrolase Gained by Fragment Screening and X-Ray Crystallography","volume":"22","year":"2014","journal-title":"Bioorg. Med. Chem"},{"key":"2026052211051083000_2026.04.18.719394v2.60","doi-asserted-by":"publisher","DOI":"10.1038\/s41589-023-01458-4"},{"key":"2026052211051083000_2026.04.18.719394v2.61","doi-asserted-by":"publisher","DOI":"10.1021\/ct400341p"},{"key":"2026052211051083000_2026.04.18.719394v2.62","doi-asserted-by":"publisher","DOI":"10.1093\/bioinformatics\/btr168"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2026.04.18.719394","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,5,22]],"date-time":"2026-05-22T18:05:20Z","timestamp":1779473120000},"score":1,"resource":{"primary":{"URL":"http:\/\/biorxiv.org\/lookup\/doi\/10.64898\/2026.04.18.719394"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,4,19]]},"references-count":62,"URL":"https:\/\/doi.org\/10.64898\/2026.04.18.719394","relation":{},"subject":[],"published":{"date-parts":[[2026,4,19]]},"subtype":"preprint"}}