{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"bioRxiv"}],"indexed":{"date-parts":[[2026,7,30]],"date-time":"2026-07-30T12:44:21Z","timestamp":1785415461059,"version":"3.56.0"},"posted":{"date-parts":[[2026,5,11]]},"group-title":"Bioinformatics","reference-count":54,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2026,5,11]],"date-time":"2026-05-11T00:00:00Z","timestamp":1778457600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/www.biorxiv.org\/about\/FAQ#license"}],"funder":[{"award":["P01AI138962"],"award-info":[{"award-number":["P01AI138962"]}],"id":[{"id":"https:\/\/ror.org\/001d55x84","id-type":"ROR","asserted-by":"publisher"}]},{"award":["U24HG007346"],"award-info":[{"award-number":["U24HG007346"]}],"id":[{"id":"https:\/\/ror.org\/001d55x84","id-type":"ROR","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2026,6,8]]},"abstract":"<jats:p>Missense variant interpretation remains a major challenge in clinical genomics. Existing missense variant impact predictors achieve strong performance, but they emphasize protein-level consequences and often share overlapping annotation priors. A missense annotation specifies the encoded amino-acid substitution, but the underlying nucleotide change may also act through nucleotide sequence context that protein-centric predictors overlook. Whether genomic language models capture distinctive nucleotide-level information beyond established missense variant impact predictors remains unclear. Through a comprehensive comparison of model backbones, embedding aggregation strategies, classifier heads, and adaptation regimes, we developed GLM-Missense, a genomic language model fine-tuned for missense variant impact prediction. Variant-position embeddings, multi-species pretraining, and low-rank adaptation were the design choices most critical to its performance. GLM-Missense contributed information complementary to established missense variant impact predictors. It showed low concordance with AlphaMissense, ESM1b, REVEL, CADD, SIFT and PolyPhen-2. We then asked whether this divergence was predictive rather than noise: after accounting for the other predictors, GLM-Missense retained the strongest unique association with variant pathogenicity. Finally, in MetaMissense, an XGBoost ensemble of all seven predictors, GLM-Missense ranked among the most informative features, indicating that its nucleotide-context signal carries information the ensemble uses to predict pathogenicity. A subset of the variants GLM-Missense resolved were annotated as missense but carried ClinVar splicing-related evidence for pathogenicity. This suggests GLM-Missense captures splicing-relevant signal. However, substituting SpliceAI for GLM-Missense in the ensemble failed to recapitulate its feature importance, suggesting that GLM-Missense captures additional sequence-derived information beyond splicing context alone. Together, these results demonstrate that GLM-Missense, a fine-tuned genomic language model, captures nucleotide-level information overlooked by established missense variant impact predictors.<\/jats:p>","DOI":"10.64898\/2026.05.06.723362","type":"posted-content","created":{"date-parts":[[2026,5,11]],"date-time":"2026-05-11T23:15:13Z","timestamp":1778541313000},"source":"Crossref","is-referenced-by-count":0,"title":["A fine-tuned genomic language model captures nucleotide-level information overlooked by missense variant impact predictors"],"prefix":"10.64898","author":[{"given":"Yaqi","family":"Su","sequence":"first","affiliation":[{"name":"University of California, Berkeley"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yu-Jen","family":"Lin","sequence":"additional","affiliation":[{"name":"University of California, Berkeley"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"54368","reference":[{"key":"2026051313101012000_2026.05.06.723362v1.1","doi-asserted-by":"publisher","DOI":"10.1038\/nmeth0410-248"},{"key":"2026051313101012000_2026.05.06.723362v1.2","doi-asserted-by":"publisher","DOI":"10.7554\/eLife.83172"},{"key":"2026051313101012000_2026.05.06.723362v1.3","doi-asserted-by":"publisher","DOI":"10.1093\/nar\/gkac1052"},{"key":"2026051313101012000_2026.05.06.723362v1.4","doi-asserted-by":"publisher","DOI":"10.1016\/j.tig.2024.11.013"},{"key":"2026051313101012000_2026.05.06.723362v1.5","unstructured":"Biderman D , Portes J , Gonzalez Ortiz JJ , et al (2024) LoRA learns less and forgets less. Transactions on Machine Learning Research URL https:\/\/openreview.net\/forum?id=aloEru2qCG"},{"key":"2026051313101012000_2026.05.06.723362v1.6","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-023-06661-w"},{"key":"2026051313101012000_2026.05.06.723362v1.7","doi-asserted-by":"publisher","DOI":"10.1038\/s41588-023-01465-0"},{"key":"2026051313101012000_2026.05.06.723362v1.8","doi-asserted-by":"publisher","DOI":"10.1007\/s00439-021-02419-4"},{"key":"2026051313101012000_2026.05.06.723362v1.9","doi-asserted-by":"publisher","DOI":"10.1145\/2939672.2939785"},{"key":"2026051313101012000_2026.05.06.723362v1.10","doi-asserted-by":"publisher","DOI":"10.1126\/science.adg7492"},{"key":"2026051313101012000_2026.05.06.723362v1.11","doi-asserted-by":"publisher","DOI":"10.1038\/nrg3046"},{"key":"2026051313101012000_2026.05.06.723362v1.12","doi-asserted-by":"publisher","DOI":"10.1038\/s10038-025-01385-3"},{"key":"2026051313101012000_2026.05.06.723362v1.13","doi-asserted-by":"publisher","DOI":"10.15252\/embr.201541116"},{"key":"2026051313101012000_2026.05.06.723362v1.14","doi-asserted-by":"publisher","DOI":"10.1038\/s41592-024-02523-z"},{"key":"2026051313101012000_2026.05.06.723362v1.15","doi-asserted-by":"publisher","DOI":"10.1126\/science.1073774"},{"key":"2026051313101012000_2026.05.06.723362v1.16","doi-asserted-by":"publisher","DOI":"10.1093\/nar\/gkh393"},{"key":"2026051313101012000_2026.05.06.723362v1.17","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-025-65823-8"},{"key":"2026051313101012000_2026.05.06.723362v1.18","doi-asserted-by":"publisher","DOI":"10.1038\/s41598-020-72404-w"},{"key":"2026051313101012000_2026.05.06.723362v1.19","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-022-31686-6"},{"key":"2026051313101012000_2026.05.06.723362v1.20","doi-asserted-by":"publisher","DOI":"10.1126\/science.185.4154.862"},{"key":"2026051313101012000_2026.05.06.723362v1.21","doi-asserted-by":"publisher","DOI":"10.1002\/humu.22768"},{"key":"2026051313101012000_2026.05.06.723362v1.22","unstructured":"Hu EJ , Shen Y , Wallis P , et al (2022) Lora: Low-rank adaptation of large language models. In: International Conference on Learning Representations, URL https:\/\/openreview.net\/forum?id=nZeVKeeFYf9"},{"key":"2026051313101012000_2026.05.06.723362v1.23","doi-asserted-by":"publisher","DOI":"10.1016\/j.ajhg.2016.08.016"},{"key":"2026051313101012000_2026.05.06.723362v1.24","doi-asserted-by":"publisher","DOI":"10.1016\/j.cell.2018.12.015"},{"key":"2026051313101012000_2026.05.06.723362v1.25","doi-asserted-by":"publisher","DOI":"10.1093\/bioinformatics\/btab083"},{"key":"2026051313101012000_2026.05.06.723362v1.26","doi-asserted-by":"publisher","DOI":"10.1038\/s41586-020-2308-7"},{"key":"2026051313101012000_2026.05.06.723362v1.27","doi-asserted-by":"publisher","DOI":"10.1038\/ng.2892"},{"key":"2026051313101012000_2026.05.06.723362v1.28","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.1611835114"},{"key":"2026051313101012000_2026.05.06.723362v1.29","doi-asserted-by":"publisher","DOI":"10.1093\/nar\/gkx1153"},{"key":"2026051313101012000_2026.05.06.723362v1.30","doi-asserted-by":"publisher","DOI":"10.1093\/nar\/gkz972"},{"key":"2026051313101012000_2026.05.06.723362v1.31","doi-asserted-by":"publisher","DOI":"10.1126\/science.ade2574"},{"key":"2026051313101012000_2026.05.06.723362v1.32","unstructured":"Liu Y , Ott M , Goyal N , et al (2019) Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:190711692"},{"key":"2026051313101012000_2026.05.06.723362v1.33","doi-asserted-by":"publisher","DOI":"10.15252\/msb.20199380"},{"key":"2026051313101012000_2026.05.06.723362v1.34","doi-asserted-by":"publisher","DOI":"10.15252\/msb.202211474"},{"key":"2026051313101012000_2026.05.06.723362v1.35","doi-asserted-by":"publisher","DOI":"10.1136\/jmedgenet-2011-100223"},{"key":"2026051313101012000_2026.05.06.723362v1.36","doi-asserted-by":"publisher","DOI":"10.1016\/S0079-7421(08)60536-8"},{"key":"2026051313101012000_2026.05.06.723362v1.37","doi-asserted-by":"publisher","DOI":"10.1186\/s12881-014-0134-1"},{"key":"2026051313101012000_2026.05.06.723362v1.38","unstructured":"Meier J , Rao R , Verkuil R , et al (2021) Language models enable zero-shot prediction of the effects of mutations on protein function. In: Ranzato M , Beygelzimer A , Dauphin Y , et al (eds) Advances in Neural Information Processing Systems, vol 34. Curran Associates, Inc., pp 29287\u201329303, URL https:\/\/proceedings.neurips.cc\/paperfiles\/paper\/2021\/file\/f51338d736f95dd42427296047067694-Paper.pdf"},{"key":"2026051313101012000_2026.05.06.723362v1.39","doi-asserted-by":"publisher","DOI":"10.1093\/nar\/gkg509"},{"key":"2026051313101012000_2026.05.06.723362v1.40","doi-asserted-by":"publisher","DOI":"10.1146\/annurev.genom.7.080505.115630"},{"key":"2026051313101012000_2026.05.06.723362v1.41","doi-asserted-by":"publisher","DOI":"10.1016\/j.cell.2015.02.029"},{"key":"2026051313101012000_2026.05.06.723362v1.42","doi-asserted-by":"publisher","DOI":"10.1007\/s00439-025-02732-2"},{"key":"2026051313101012000_2026.05.06.723362v1.43","doi-asserted-by":"publisher","DOI":"10.1093\/nar\/gky1016"},{"key":"2026051313101012000_2026.05.06.723362v1.44","doi-asserted-by":"publisher","DOI":"10.1038\/gim.2015.30"},{"key":"2026051313101012000_2026.05.06.723362v1.45","doi-asserted-by":"publisher","DOI":"10.1002\/humu.22273"},{"key":"2026051313101012000_2026.05.06.723362v1.46","unstructured":"Schiff Y , Kao CH , Gokaslan A , et al (2024) Caduceus: Bi-directional equivariant long-range DNA sequence modeling. In: Proceedings of the 41st International Conference on Machine Learning, Proceedings of Machine Learning Research, vol 235. PMLR, pp 43632\u201343648, URL https:\/\/proceedings.mlr.press\/v235\/schiff24a.html"},{"key":"2026051313101012000_2026.05.06.723362v1.47","doi-asserted-by":"publisher","DOI":"10.1038\/ng.3837"},{"key":"2026051313101012000_2026.05.06.723362v1.48","doi-asserted-by":"publisher","DOI":"10.1016\/j.ajhg.2017.07.014"},{"key":"2026051313101012000_2026.05.06.723362v1.49","doi-asserted-by":"publisher","DOI":"10.1016\/j.jmb.2013.07.014"},{"key":"2026051313101012000_2026.05.06.723362v1.50","doi-asserted-by":"publisher","DOI":"10.1126\/science.1243490"},{"key":"2026051313101012000_2026.05.06.723362v1.51","doi-asserted-by":"publisher","DOI":"10.1126\/science.1219240"},{"key":"2026051313101012000_2026.05.06.723362v1.52","doi-asserted-by":"publisher","DOI":"10.1038\/s44320-025-00184-4"},{"key":"2026051313101012000_2026.05.06.723362v1.53","doi-asserted-by":"publisher","DOI":"10.7554\/eLife.45396"},{"key":"2026051313101012000_2026.05.06.723362v1.54","doi-asserted-by":"publisher","DOI":"10.1126\/science.1254806"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2026.05.06.723362","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,9]],"date-time":"2026-06-09T07:30:13Z","timestamp":1780990213000},"score":1,"resource":{"primary":{"URL":"http:\/\/biorxiv.org\/lookup\/doi\/10.64898\/2026.05.06.723362"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,5,11]]},"references-count":54,"URL":"https:\/\/doi.org\/10.64898\/2026.05.06.723362","relation":{},"subject":[],"published":{"date-parts":[[2026,5,11]]},"subtype":"preprint"}}