{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"bioRxiv"}],"indexed":{"date-parts":[[2026,7,30]],"date-time":"2026-07-30T16:56:02Z","timestamp":1785430562287,"version":"3.56.0"},"posted":{"date-parts":[[2026,6,2]]},"group-title":"Bioinformatics","reference-count":35,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2026,6,2]],"date-time":"2026-06-02T00:00:00Z","timestamp":1780358400000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by-nc\/4.0\/"}],"funder":[{"award":["R01CA262296"],"award-info":[{"award-number":["R01CA262296"]}],"id":[{"id":"https:\/\/ror.org\/040gcmg81","id-type":"ROR","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2026,6,2]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                <jats:sec>\n                  <jats:title>Motivation<\/jats:title>\n                  <jats:p>Models deployed for genomic prediction of diseases perform unevenly across populations, limiting clinical utility. Two factors drive this limitation: large imbalances in sample availability across ancestry groups and non-stationarity of genotype\u2013phenotype effect sizes across the ancestry continuum. While tabular foundation models with in-context learning (ICL) have shown strong sample efficiency in other domains, their effectiveness for genotype-to-phenotype prediction and their robustness to ancestry-driven effect heterogeneity remain unclear.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Results<\/jats:title>\n                  <jats:p>Using large, ancestrally diverse biobank data, we show that ICL-capable tabular foundation models reduce performance degradation in under-sampled ancestry groups compared to conventional supervised approaches. However, we find that prevailing models trained on existing synthetic tabular tasks fail when allele effect sizes vary across ancestry space. Treating genetic ancestry as a continuous variable, we introduce an instruction-tuning framework that exposes models to synthetic tasks with ancestry-dependent non-stationary effects. Instruction-tuned models achieve improved and more stable predictive performance across the genetic ancestry continuum, including for individuals distant from in-context exemplars in ancestry space.<\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Availability and Implementation<\/jats:title>\n                  <jats:p>\n                    All code for instruction-tuning models, synthetic task generation, data wrangling, and model evaluation, is publicly available at\n                    <jats:ext-link xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" ext-link-type=\"uri\" xlink:href=\"https:\/\/github.com\/ai4pm\/Bridging-Ancestry-Gaps-in-Genomic-Risk-Prediction-with-Tabular-Foundation-Models\">https:\/\/github.com\/ai4pm\/Bridging-Ancestry-Gaps-in-Genomic-Risk-Prediction-with-Tabular-Foundation-Models<\/jats:ext-link>\n                    . The final instruction-tuned model (ICL-NS-G2P-proto) is also released in this repository. Detailed documentation is provided, including environment setup instructions and guidelines for running various parts. The instruction-tuning task datasets are available at\n                    <jats:ext-link xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" ext-link-type=\"uri\" xlink:href=\"https:\/\/zenodo.org\/records\/18309187\">https:\/\/zenodo.org\/records\/18309187<\/jats:ext-link>\n                    .\n                  <\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Contact<\/jats:title>\n                  <jats:p>\n                    <jats:email>adas23@uthsc.edu<\/jats:email>\n                    ,\n                    <jats:email>ycui2@uthsc.edu<\/jats:email>\n                  <\/jats:p>\n                <\/jats:sec>\n                <jats:sec>\n                  <jats:title>Supplementary Information<\/jats:title>\n                  <jats:p>Supplementary data are available online.<\/jats:p>\n                <\/jats:sec>","DOI":"10.64898\/2026.05.29.728877","type":"posted-content","created":{"date-parts":[[2026,6,2]],"date-time":"2026-06-02T21:55:12Z","timestamp":1780437312000},"source":"Crossref","is-referenced-by-count":0,"title":["Bridging Ancestry Gaps in Genomic Risk Prediction with Tabular Foundation Models"],"prefix":"10.64898","author":[{"given":"Anirban","family":"Das","sequence":"first","affiliation":[{"name":"Department of Genetics, Genomics & Informatics, The University of Tennessee Health Science Center, Memphis, 38163, Tennessee, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-7577-6845","authenticated-orcid":false,"given":"Yan","family":"Cui","sequence":"additional","affiliation":[{"name":"Department of Genetics, Genomics & Informatics, The University of Tennessee Health Science Center, Memphis, 38163, Tennessee, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"54368","reference":[{"key":"2026060409351650000_2026.05.29.728877v1.1","doi-asserted-by":"crossref","unstructured":"Y. Bai , C. Fan , H. Wang , C. Xiong , and S. Mei . Transformers as statisticians: Provable in-context learning with in-context algorithm selection. In Advances in Neural Information Processing Systems, 2023.","DOI":"10.52202\/075280-2498"},{"key":"2026060409351650000_2026.05.29.728877v1.2","doi-asserted-by":"crossref","unstructured":"S. A. Bien , G. L. Wojcik , C. J. Hodonsky , C. R. Gignoux , I. Cheng , T. C. Matise , U. Peters , E. E. Kenny , and K. E. North . The future of genomic studies must be globally representative: Perspectives from page. Annual Review of Genomics and Human Genetics, 2019.","DOI":"10.1146\/annurev-genom-091416-035517"},{"key":"2026060409351650000_2026.05.29.728877v1.3","first-page":"1877","article-title":"Language models are few-shot learners","volume":"33","year":"2020","journal-title":"Advances in Neural Information Processing Systems"},{"key":"2026060409351650000_2026.05.29.728877v1.4","first-page":"18878","article-title":"Data distributional properties drive emergent in-context learning in transformers","volume":"35","year":"2022","journal-title":"Advances in Neural Information Processing Systems"},{"key":"2026060409351650000_2026.05.29.728877v1.5","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-023-06957-x"},{"key":"2026060409351650000_2026.05.29.728877v1.6","doi-asserted-by":"publisher","DOI":"10.1038\/s41596-020-0353-1"},{"key":"2026060409351650000_2026.05.29.728877v1.7","doi-asserted-by":"publisher","DOI":"10.1038\/s41746-025-01564-8"},{"key":"2026060409351650000_2026.05.29.728877v1.8","unstructured":"A. Das , M. I. Khalid , R. Pe\u00f1aloza , and S. Schockaert . When no paths lead to rome: Benchmarking systematic neural relational reasoning. Advances in Neural Information Processing Systems, 39, 2025."},{"key":"2026060409351650000_2026.05.29.728877v1.9","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-023-06079-4"},{"key":"2026060409351650000_2026.05.29.728877v1.10","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-019-11112-0"},{"key":"2026060409351650000_2026.05.29.728877v1.11","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-020-18918-3"},{"key":"2026060409351650000_2026.05.29.728877v1.12","doi-asserted-by":"publisher","DOI":"10.1186\/s13073-024-01345-0"},{"key":"2026060409351650000_2026.05.29.728877v1.13","doi-asserted-by":"publisher","DOI":"10.1146\/annurev-biodatasci-020722-020704"},{"key":"2026060409351650000_2026.05.29.728877v1.14","doi-asserted-by":"publisher","DOI":"10.1038\/s41598-018-32264-x"},{"key":"2026060409351650000_2026.05.29.728877v1.15","doi-asserted-by":"publisher","DOI":"10.1038\/s41576-019-0144-0"},{"key":"2026060409351650000_2026.05.29.728877v1.16","doi-asserted-by":"publisher","DOI":"10.1038\/s41588-024-02064-3"},{"key":"2026060409351650000_2026.05.29.728877v1.17","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-024-08328-6"},{"key":"2026060409351650000_2026.05.29.728877v1.18","doi-asserted-by":"publisher","DOI":"10.1038\/s41576-023-00637-2"},{"key":"2026060409351650000_2026.05.29.728877v1.19","doi-asserted-by":"publisher","DOI":"10.1038\/s41576-025-00900-8"},{"key":"2026060409351650000_2026.05.29.728877v1.20","doi-asserted-by":"publisher","DOI":"10.1038\/s41588-025-02140-2"},{"key":"2026060409351650000_2026.05.29.728877v1.21","doi-asserted-by":"publisher","DOI":"10.1016\/j.ajhg.2022.05.002"},{"key":"2026060409351650000_2026.05.29.728877v1.22","doi-asserted-by":"publisher","DOI":"10.1016\/j.tig.2021.06.004"},{"issue":"1","key":"2026060409351650000_2026.05.29.728877v1.23","first-page":"49","article-title":"On the generalised distance in statistics","volume":"2","year":"1936","journal-title":"Proceedings of the National Institute of Sciences of India"},{"key":"2026060409351650000_2026.05.29.728877v1.24","doi-asserted-by":"publisher","DOI":"10.1038\/s41588-020-0580-y"},{"key":"2026060409351650000_2026.05.29.728877v1.25","doi-asserted-by":"publisher","DOI":"10.1186\/s40246-024-00664-y"},{"key":"2026060409351650000_2026.05.29.728877v1.26","doi-asserted-by":"publisher","DOI":"10.1016\/j.cell.2019.08.051"},{"key":"2026060409351650000_2026.05.29.728877v1.27","doi-asserted-by":"publisher","DOI":"10.1016\/j.ajhg.2021.11.008"},{"key":"2026060409351650000_2026.05.29.728877v1.28","doi-asserted-by":"crossref","unstructured":"T. Sharma , N. K. Verma , and Y. Cui . Omics-based cancer prognosis across ethnic groups: From feature engineering to disparity detection and mitigation. IEEE Transactions on Artificial Intelligence, 2026. In press.","DOI":"10.1109\/TAI.2025.3640596"},{"key":"2026060409351650000_2026.05.29.728877v1.29","doi-asserted-by":"publisher","DOI":"10.1038\/s41588-023-01314-0"},{"key":"2026060409351650000_2026.05.29.728877v1.30","doi-asserted-by":"publisher","DOI":"10.1016\/j.cell.2019.02.048"},{"key":"2026060409351650000_2026.05.29.728877v1.31","doi-asserted-by":"publisher","DOI":"10.1038\/s41591-022-01891-3"},{"key":"2026060409351650000_2026.05.29.728877v1.32","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-020-17719-y"},{"key":"2026060409351650000_2026.05.29.728877v1.33","doi-asserted-by":"publisher","DOI":"10.1016\/j.xgen.2023.100408"},{"key":"2026060409351650000_2026.05.29.728877v1.34","unstructured":"J. Wei , M. Bosma , V. Y. Zhao , K. Guu , A. W. Yu , B. Lester , N. Du , A. M. Dai , and Q. V. Le . Finetuned language models are zero-shot learners. In International Conference on Learning Representations, 2022."},{"key":"2026060409351650000_2026.05.29.728877v1.35","doi-asserted-by":"publisher","DOI":"10.1016\/j.xgen.2022.100192"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2026.05.29.728877","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,4]],"date-time":"2026-06-04T16:35:28Z","timestamp":1780590928000},"score":1,"resource":{"primary":{"URL":"http:\/\/biorxiv.org\/lookup\/doi\/10.64898\/2026.05.29.728877"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,6,2]]},"references-count":35,"URL":"https:\/\/doi.org\/10.64898\/2026.05.29.728877","relation":{},"subject":[],"published":{"date-parts":[[2026,6,2]]},"subtype":"preprint"}}