{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,28]],"date-time":"2026-08-28T18:16:18Z","timestamp":1787940978070,"version":"build-2784847793"},"reference-count":28,"publisher":"Oxford University Press (OUP)","issue":"8","license":[{"start":{"date-parts":[[2026,8,6]],"date-time":"2026-08-06T00:00:00Z","timestamp":1785974400000},"content-version":"vor","delay-in-days":5,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/100000002","name":"National Institutes of Health","doi-asserted-by":"publisher","award":["1R35GM158529"],"award-info":[{"award-number":["1R35GM158529"]}],"id":[{"id":"10.13039\/100000002","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2026,8,3]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:sec>\n                    <jats:title>Motivation<\/jats:title>\n                    <jats:p>Multimodal learning is often assumed to improve predictive performance by combining complementary views, yet in biological applications auxiliary modalities are frequently imperfect, incomplete, or derived from upstream predictors and heuristics. We study this issue in TCR\u2013peptide binding prediction, where sequence embeddings from pretrained protein language models are strong and transferable, but structure-derived residue graphs must be built from predicted folds and discretized contacts. These structural views can therefore be noisy, inconsistent across proteins, and sensitive to modeling choices, making them difficult to optimize jointly with sequence features. In this setting, naive sequence\u2009+\u2009graph fusion can destabilize training and degrade generalization, falling below a sequence-only baseline when supervision is scarce or contacts are noisy. This motivates a practical goal: use imperfect structural information when it helps, without sacrificing stability when it does not.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>We introduce TRACE, a lightweight framework that encodes each entity (TCR and peptide) with parallel sequence (frozen ESM-2) and residue-graph (GNN) towers, then applies CLIP-style intra-entity contrastive alignment before interaction modeling. The alignment encourages modality-consistent representations for the same biological entity, preventing noisy graph signals from dominating fusion. We evaluate under a leakage-controlled TCHard RN protocol with pair-disjoint splits, training-only model selection, and exclusion of negative-sampling metadata that otherwise trivially inflates AUROC. In this setting the task is near chance for all methods, and we do not claim state-of-the-art accuracy; our contribution is the failure-mode analysis, the alignment stabilizer, and the audited protocol itself. Among matched baselines TRACE attains the best mean AUROC (0.578\u00b10.033 over five folds), and an ablation shows that intra-entity alignment acts as a stabilizer: it gives a small but consistent full-data gain (better on 4 of 5 folds), stays robust under substantial graph-edge corruption, and prevents collapse toward chance under limited supervision (+0.05 AUROC at 10%\u201320% of labels), the regime where unconstrained fusion fails. How modalities are integrated, and how carefully they are evaluated, matters more than how many are used.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Availability and implementation<\/jats:title>\n                    <jats:p>Code is available at https:\/\/github.com\/MineSelf2016\/TRACE and archived at https:\/\/doi.org\/10.5281\/zenodo.20635593. Data are available at https:\/\/doi.org\/10.6084\/m9.figshare.31991007.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.1093\/bioinformatics\/btag573","type":"journal-article","created":{"date-parts":[[2026,8,6]],"date-time":"2026-08-06T11:59:00Z","timestamp":1786017540000},"source":"Crossref","is-referenced-by-count":0,"title":["When multimodal fusion fails: contrastive alignment as a necessary stabilizer for TCR\u2013peptide binding prediction"],"prefix":"10.1093","volume":"42","author":[{"ORCID":"https:\/\/orcid.org\/0009-0001-8940-0996","authenticated-orcid":false,"given":"Cong","family":"Qi","sequence":"first","affiliation":[{"name":"Department of Computer Science, New Jersey Institute of Technology , Newark, NJ 07102,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Wenbo","family":"Wang","sequence":"additional","affiliation":[{"name":"Department of Computer Science, Hamilton College , Clinton, NY 13323,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Hanzhang","family":"Fang","sequence":"additional","affiliation":[{"name":"Department of Computer Science, New Jersey Institute of Technology , Newark, NJ 07102,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-6059-4267","authenticated-orcid":false,"given":"Zhi","family":"Wei","sequence":"additional","affiliation":[{"name":"Department of Computer Science, New Jersey Institute of Technology , Newark, NJ 07102,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"286","published-online":{"date-parts":[[2026,8,6]]},"reference":[{"key":"2026082813175888900_btag573-B1","doi-asserted-by":"publisher","first-page":"3272","DOI":"10.1109\/tcbb.2021.3115353","article-title":"An attention based bidirectional LSTM method to predict the binding of TCR and epitope","volume":"19","author":"Bi","year":"2022","journal-title":"IEEE\/ACM Trans Comput Biol Bioinform"},{"key":"2026082813175888900_btag573-B2","author":"Bolouri","year":"2025"},{"key":"2026082813175888900_btag573-B3","doi-asserted-by":"publisher","first-page":"e0324011","DOI":"10.1371\/journal.pone.0324011","article-title":"Assessing the generalization capabilities of TCR binding predictors via peptide distance analysis","volume":"20","author":"Castorina","year":"2025","journal-title":"PLoS One"},{"key":"2026082813175888900_btag573-B4","author":"Chen","year":"2020"},{"key":"2026082813175888900_btag573-B5","doi-asserted-by":"publisher","first-page":"1128326","DOI":"10.3389\/fimmu.2023.1128326","article-title":"Performance comparison of TCR-PMHC prediction tools reveals a strong data dependency","volume":"14","author":"Deng","year":"2023","journal-title":"Front Immunol"},{"key":"2026082813175888900_btag573-B6","doi-asserted-by":"publisher","first-page":"1060","DOI":"10.1038\/s42256-023-00727-0","article-title":"The pitfalls of negative data bias for the t-cell epitope specificity challenge","volume":"5","author":"Dens","year":"2023","journal-title":"Nat Mach Intell"},{"key":"2026082813175888900_btag573-B7","doi-asserted-by":"crossref","first-page":"e1013679","DOI":"10.1371\/journal.pcbi.1013679","article-title":"Oneprot: towards multi-modal protein foundation models via latent space alignment of sequence, structure, binding sites and text encoders","volume":"21","author":"Fl\u00f6ge","year":"2025","journal-title":"PLoS Comput Biol"},{"key":"2026082813175888900_btag573-B8","author":"Gilmer","year":"2017"},{"key":"2026082813175888900_btag573-B9","author":"Hamilton","year":"2017"},{"key":"2026082813175888900_btag573-B10","doi-asserted-by":"publisher","author":"Jiang Y, Huo M, Li SC","DOI":"10.1093\/bib\/bbad086"},{"key":"2026082813175888900_btag573-B11","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pcbi.1008814"},{"key":"2026082813175888900_btag573-B12","doi-asserted-by":"crossref","first-page":"583","DOI":"10.1038\/s41586-021-03819-2","article-title":"Highly accurate protein structure prediction with alphafold","volume":"596","author":"Jumper","year":"2021","journal-title":"Nature"},{"key":"2026082813175888900_btag573-B13","doi-asserted-by":"publisher","author":"Jurtz","year":"2018","DOI":"10.1101\/433706"},{"key":"2026082813175888900_btag573-B14"},{"key":"2026082813175888900_btag573-B15","doi-asserted-by":"crossref","first-page":"1145","DOI":"10.1021\/acs.jcim.3c01753","article-title":"Coati: multimodal contrastive pretraining for representing and traversing chemical space","volume":"64","author":"Kaufman","year":"2024","journal-title":"J Chem Inf Model"},{"key":"2026082813175888900_btag573-B16","doi-asserted-by":"publisher","DOI":"10.1093\/bioinformatics\/btae472"},{"key":"2026082813175888900_btag573-B17","author":"Kipf","year":"2017"},{"key":"2026082813175888900_btag573-B18","doi-asserted-by":"publisher","first-page":"1346784","DOI":"10.3389\/fgene.2024.1346784","article-title":"Tcrcost: a deep learning model utilizing TCR 3D structure for enhanced of TCR\u2013peptide binding","volume":"15","author":"Li","year":"2024","journal-title":"Front Genet"},{"key":"2026082813175888900_btag573-B19","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pcbi.1013050"},{"key":"2026082813175888900_btag573-B20","doi-asserted-by":"publisher","first-page":"1060","DOI":"10.1038\/s42003-021-02610-3","article-title":"Nettcr-2.0 enables accurate prediction of TCR-peptide binding by using paired TCR \u03b1 and \u03b2 sequence data","volume":"4","author":"Montemurro","year":"2021","journal-title":"Commun Biol"},{"key":"2026082813175888900_btag573-B21","doi-asserted-by":"publisher","DOI":"10.1093\/bioinformatics\/btad284","article-title":"epitcr: a highly sensitive predictor for TCR\u2013peptide binding","volume":"39","author":"Nguyen Pham","year":"2023","journal-title":"Bioinformatics"},{"key":"2026082813175888900_btag573-B22","author":"Radford","year":"2021"},{"key":"2026082813175888900_btag573-B23","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-021-21879-w"},{"key":"2026082813175888900_btag573-B24","doi-asserted-by":"publisher","first-page":"664514","DOI":"10.3389\/fimmu.2021.664514","article-title":"Contribution of t cell receptor alpha and beta cdr3, MHC typing, v and j genes to peptide binding prediction","volume":"12","author":"Springer","year":"2021","journal-title":"Front Immunol"},{"key":"2026082813175888900_btag573-B25","doi-asserted-by":"publisher","author":"van den Oord","year":"2018","DOI":"10.48550\/arXiv.1807.03748"},{"key":"2026082813175888900_btag573-B26","author":"Veli\u010dkovi\u0107","year":"2018"},{"key":"2026082813175888900_btag573-B27","first-page":"5083","author":"Wang","year":"2025"},{"key":"2026082813175888900_btag573-B28","doi-asserted-by":"publisher","first-page":"i237","DOI":"10.1093\/bioinformatics\/btab294","article-title":"Titan: T-cell receptor specificity prediction with bimodal attention networks","volume":"37","author":"Weber","year":"2021","journal-title":"Bioinformatics"}],"container-title":["Bioinformatics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/academic.oup.com\/bioinformatics\/advance-article-pdf\/doi\/10.1093\/bioinformatics\/btag573\/70507017\/btag573.pdf","content-type":"application\/pdf","content-version":"am","intended-application":"syndication"},{"URL":"https:\/\/academic.oup.com\/bioinformatics\/article-pdf\/42\/8\/btag573\/70507017\/btag573.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/academic.oup.com\/bioinformatics\/article-pdf\/42\/8\/btag573\/70507017\/btag573.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,8,28]],"date-time":"2026-08-28T17:18:05Z","timestamp":1787937485000},"score":1,"resource":{"primary":{"URL":"https:\/\/academic.oup.com\/bioinformatics\/article\/doi\/10.1093\/bioinformatics\/btag573\/8753144"}},"subtitle":[],"editor":[{"given":"Daisuke","family":"Kihara","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"editor"}]}],"short-title":[],"issued":{"date-parts":[[2026,8]]},"references-count":28,"journal-issue":{"issue":"8","published-print":{"date-parts":[[2026,8,3]]}},"URL":"https:\/\/doi.org\/10.1093\/bioinformatics\/btag573","relation":{},"ISSN":["1367-4803","1367-4811"],"issn-type":[{"value":"1367-4803","type":"print"},{"value":"1367-4811","type":"electronic"}],"subject":[],"published-other":{"date-parts":[[2026,8]]},"published":{"date-parts":[[2026,8]]},"article-number":"btag573"}}