{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"institution":[{"name":"bioRxiv"}],"indexed":{"date-parts":[[2026,9,28]],"date-time":"2026-09-28T21:03:06Z","timestamp":1790629386592,"version":"4.1.0"},"posted":{"date-parts":[[2025,12,13]]},"group-title":"Bioinformatics","reference-count":49,"publisher":"openRxiv","license":[{"start":{"date-parts":[[2025,12,13]],"date-time":"2025-12-13T00:00:00Z","timestamp":1765584000000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by-nc\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"accepted":{"date-parts":[[2026,9,24]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                <jats:p>Genomic language models (gLMs) are powerful numerical surrogates for DNA, but existing architectures focus primarily on human DNA or limited viral sets, and no dedicated benchmark exists for viral genome understanding. Here we introduce Vir2vec (17M\u2013422M parameters), decoder-only gLMs continually pretrained on a curated pan-viral corpus of 565,747 complete genomes across 295 species. We also present the viral Genome Understanding Evaluation (vGUE), a benchmark probing viral representations across broad organism discrimination, evolutionary signatures, intra-genus separation, variant subtyping, and host tropism. Evaluated as frozen feature extractors under nested cross-validation, Vir2vec achieves top balanced accuracy across 5 of 7 vGUE tasks, outperforming human-trained (Mistral-DNA), viral-specific (ModernBERT-virus), and broad biological foundation (Evo-1) model. Furthermore, Vir2vec displays robust zero-shot generalization to unseen viral families. Finally, we introduce an argmax-routed SHAP explainability framework that projects feature attributions directly onto single-codon coordinates, correctly isolating key receptor-binding domain mutations driving mouse host adaptation. Vir2vec and vGUE provide a foundation for viral surveillance and discovery pipelines.<\/jats:p>\n                <jats:sec>\n                  <jats:title>Key Messages<\/jats:title>\n                  <jats:list list-type=\"bullet\">\n                    <jats:list-item>\n                      <jats:p>We present Vir2vec, a family of decoder-only viral genomic language models (17M, 138M, and 422M parameters) obtained via continual pretraining of Mistral-DNA on a curated pan-viral corpus (565,747 complete genomes across 295 species), generating fixed-length 4,096-dimensional embeddings for whole genomes and reads.<\/jats:p>\n                    <\/jats:list-item>\n                    <jats:list-item>\n                      <jats:p>We introduce vGUE, a dedicated evaluation suite for viral genomic foundation models spanning five biological dimensions: broad organism-level distinctions, genome-wide evolutionary fingerprints, intra-genus separation, lineage\/subtype typing, and phenotypic context signal detection.<\/jats:p>\n                    <\/jats:list-item>\n                    <jats:list-item>\n                      <jats:p>Under a unified, leakage-controlled evaluation protocol (frozen embeddings plus shallow classifiers under nested cross-validation), Vir2vec consistently outperforms human-trained (Mistral-DNA), viral-centric (ModernBERT-virus), and 7B-parameter prokaryotic\/phage (Evo-1) baselines across genome-wide viral tasks.<\/jats:p>\n                    <\/jats:list-item>\n                    <jats:list-item>\n                      <jats:p>\n                        Pretraining data ablations demonstrate robust zero-shot out-of-distribution generalization to unseen viral taxa (\n                        <jats:italic>&gt;<\/jats:italic>\n                        97% balanced accuracy on completely held-out\n                        <jats:italic>Enterovirus A<\/jats:italic>\n                        and\n                        <jats:italic>Picornaviridae<\/jats:italic>\n                        ), confirming that Vir2vec internalizes universal viral sequence syntax.\n                      <\/jats:p>\n                    <\/jats:list-item>\n                    <jats:list-item>\n                      <jats:p>We introduce an argmax-routed SHAP explainability framework that projects latent feature attributions back to single-codon resolution, validating that the model autonomously pinpoints key receptor-binding domain mutations (such as codon N501 in SARS-CoV-2) driving host adaptation.<\/jats:p>\n                    <\/jats:list-item>\n                    <jats:list-item>\n                      <jats:p>Context-truncated short reads (150-bp) and subtle intra-host tissue tropisms remain primary performance bottlenecks, while compact Vir2vec scale variants (17M and 138M) capture nearly the entire pan-viral latent manifold, offering lightweight options for resource-constrained edge deployment.<\/jats:p>\n                    <\/jats:list-item>\n                  <\/jats:list>\n                <\/jats:sec>","DOI":"10.64898\/2025.12.12.693901","type":"posted-content","created":{"date-parts":[[2025,12,14]],"date-time":"2025-12-14T06:15:16Z","timestamp":1765692916000},"source":"Crossref","is-referenced-by-count":1,"title":["Vir2vec: A Genome-Wide Viral Embedding"],"prefix":"10.64898","author":[{"ORCID":"https:\/\/orcid.org\/0009-0002-4405-1697","authenticated-orcid":false,"given":"Simone","family":"Rancati","sequence":"first","affiliation":[{"name":"Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Via Adolfo Ferrata 5, 27100, Pavia, Italy"},{"name":"Department of Epidemiology, College of Public Health and Health Professions, University of Florida, 2004 Mowry Road, 32610, Gainesville, FL, United States"},{"name":"Emerging Pathogens Institute, University of Florida, 2055 Mowry Road, 32610, Gainesville, FL, United States"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0002-3188-2504","authenticated-orcid":false,"given":"Pablo","family":"Arozarena Donelli","sequence":"additional","affiliation":[{"name":"Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Via Adolfo Ferrata 5, 27100, Pavia, Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-7007-0862","authenticated-orcid":false,"given":"Giovanna","family":"Nicora","sequence":"additional","affiliation":[{"name":"Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Via Adolfo Ferrata 5, 27100, Pavia, Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0006-0359-5128","authenticated-orcid":false,"given":"Laura","family":"Bergomi","sequence":"additional","affiliation":[{"name":"Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Via Adolfo Ferrata 5, 27100, Pavia, Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-2887-088X","authenticated-orcid":false,"given":"Tommaso Mario","family":"Buonocore","sequence":"additional","affiliation":[{"name":"Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Via Adolfo Ferrata 5, 27100, Pavia, Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9281-7592","authenticated-orcid":false,"given":"Micheal Aaron","family":"Sy","sequence":"additional","affiliation":[{"name":"Department of Epidemiology, College of Public Health and Health Professions, University of Florida, 2004 Mowry Road, 32610, Gainesville, FL, United States"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sakshi","family":"Pandey","sequence":"additional","affiliation":[{"name":"Department of Computer and Information Science and Engineering, University of Florida, 432 Newell Dr, 32611, Gainesville, FL, United States"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-9021-5595","authenticated-orcid":false,"given":"Mattia","family":"Prosperi","sequence":"additional","affiliation":[{"name":"Department of Epidemiology, College of Public Health and Health Professions, University of Florida, 2004 Mowry Road, 32610, Gainesville, FL, United States"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Marco","family":"Salemi","sequence":"additional","affiliation":[{"name":"Emerging Pathogens Institute, University of Florida, 2055 Mowry Road, 32610, Gainesville, FL, United States"},{"name":"Department of Pathology, Immunology and Laboratory Medicine, College of Medicine, University of Florida, 1600 SW Archer Road, 32610, Gainesville, FL, United States"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6974-9808","authenticated-orcid":false,"given":"Riccardo","family":"Bellazzi","sequence":"additional","affiliation":[{"name":"Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Via Adolfo Ferrata 5, 27100, Pavia, Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Christina","family":"Boucher","sequence":"additional","affiliation":[{"name":"Department of Computer and Information Science and Engineering, University of Florida, 432 Newell Dr, 32611, Gainesville, FL, United States"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0679-828X","authenticated-orcid":false,"given":"Enea","family":"Parimbelli","sequence":"additional","affiliation":[{"name":"Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Via Adolfo Ferrata 5, 27100, Pavia, Italy"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-5704-3533","authenticated-orcid":false,"given":"Simone","family":"Marini","sequence":"additional","affiliation":[{"name":"Department of Epidemiology, College of Public Health and Health Professions, University of Florida, 2004 Mowry Road, 32610, Gainesville, FL, United States"},{"name":"Emerging Pathogens Institute, University of Florida, 2055 Mowry Road, 32610, Gainesville, FL, United States"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"54368","reference":[{"key":"2026092813251460000_2025.12.12.693901v4.1","unstructured":"World Health Organization. Influenza (seasonal). https:\/\/www.who.int\/news-room\/fact-sheets\/detail\/influenza-%28seasonal%29, February 2025. Accessed: 18 November 2025."},{"key":"2026092813251460000_2025.12.12.693901v4.2","unstructured":"World Health Organization. Dengue and severe dengue. https:\/\/www.who.int\/news-room\/fact-sheets\/detail\/dengue-and-severe-dengue, August 2025. Accessed: 18 November 2025."},{"key":"2026092813251460000_2025.12.12.693901v4.3","doi-asserted-by":"crossref","first-page":"15955","DOI":"10.1038\/ncomms15955","article-title":"Marine viruses discovered via metagenomics shed light on viral strategies throughout the oceans","volume":"8","year":"2017","journal-title":"Nature Communications"},{"key":"2026092813251460000_2025.12.12.693901v4.4","doi-asserted-by":"crossref","first-page":"12","DOI":"10.1186\/s13100-017-0095-y","article-title":"Viral communities of the human gut: metagenomic analysis of composition and dynamics","volume":"8","year":"2017","journal-title":"Mobile DNA"},{"issue":"3","key":"2026092813251460000_2025.12.12.693901v4.5","doi-asserted-by":"crossref","first-page":"699","DOI":"10.1053\/j.gastro.2022.05.048","article-title":"Uncovering 1,058 novel human enteric DNA viruses through deep long-read third-generation sequencing and their clinical impact","volume":"163","year":"2022","journal-title":"Gastroenterology"},{"key":"2026092813251460000_2025.12.12.693901v4.6","doi-asserted-by":"crossref","first-page":"1605","DOI":"10.1016\/j.csbj.2020.06.019","article-title":"Computational approaches in viral ecology","volume":"18","year":"2020","journal-title":"Computational and Structural Biotechnology Journal"},{"key":"2026092813251460000_2025.12.12.693901v4.7","doi-asserted-by":"publisher","DOI":"10.1128\/msystems.01105-23"},{"issue":"2","key":"2026092813251460000_2025.12.12.693901v4.8","doi-asserted-by":"crossref","first-page":"278","DOI":"10.3390\/v14020278","article-title":"The human virome: viral metagenomics, relations with human diseases, and therapeutic applications","volume":"14","year":"2022","journal-title":"Viruses"},{"key":"2026092813251460000_2025.12.12.693901v4.9","doi-asserted-by":"crossref","unstructured":"Moreno Zolfo et al. Discovering and exploring the hidden diversity of human gut viruses using highly enriched virome samples. bioRxiv, page 2024.02.19.580813, 2024.","DOI":"10.1101\/2024.02.19.580813"},{"key":"2026092813251460000_2025.12.12.693901v4.10","unstructured":"Garyk Brixi et al. Genome modeling and design across all domains of life with Evo 2. bioRxiv, page 2025.02.18.638918, 2025."},{"issue":"4","key":"2026092813251460000_2025.12.12.693901v4.11","doi-asserted-by":"crossref","first-page":"bbaf384","DOI":"10.1093\/bib\/bbaf384","article-title":"SARITA: a large language model for generating the S1 subunit of the SARS-CoV-2 spike protein","volume":"26","year":"2025","journal-title":"Briefings in Bioinformatics"},{"key":"2026092813251460000_2025.12.12.693901v4.12","doi-asserted-by":"crossref","first-page":"287","DOI":"10.1038\/s41592-024-02523-z","article-title":"Nucleotide transformer: building and evaluating robust foundation models for human genomics","volume":"22","year":"2025","journal-title":"Nature Methods"},{"key":"2026092813251460000_2025.12.12.693901v4.13","article-title":"DNABERT-2: Efficient foundation model and benchmark for multi-species genome","year":"2024","journal-title":"arXiv preprint"},{"issue":"1","key":"2026092813251460000_2025.12.12.693901v4.14","doi-asserted-by":"crossref","first-page":"vbae117","DOI":"10.1093\/bioadv\/vbae117","article-title":"Understanding the natural language of DNA using encoder-decoder foundation models with bytelevel precision","volume":"4","year":"2024","journal-title":"Bioinformatics Advances"},{"key":"2026092813251460000_2025.12.12.693901v4.15","article-title":"Mixtral of experts","year":"2024","journal-title":"arXiv preprint"},{"key":"2026092813251460000_2025.12.12.693901v4.16","unstructured":"Raphael Mourad. Mistral-DNA-v1-422M-hg38. Hugging Face. Online resource."},{"key":"2026092813251460000_2025.12.12.693901v4.17","unstructured":"Raphael Mourad. ModernBert-DNA-v1-37M-virus. Hugging Face. Online resource."},{"key":"2026092813251460000_2025.12.12.693901v4.18","article-title":"Smarter, better, faster, longer: A modern bidirectional encoder for fast, memory efficient, and long-context finetuning and inference","year":"2024","journal-title":"arXiv preprint"},{"key":"2026092813251460000_2025.12.12.693901v4.19","doi-asserted-by":"publisher","DOI":"10.1093\/bioinformatics\/btae704"},{"issue":"6","key":"2026092813251460000_2025.12.12.693901v4.20","doi-asserted-by":"crossref","first-page":"302","DOI":"10.3390\/a18060302","article-title":"Enhanced viral genome classification using large language models","volume":"18","year":"2025","journal-title":"Algorithms"},{"key":"2026092813251460000_2025.12.12.693901v4.21","doi-asserted-by":"crossref","first-page":"3547","DOI":"10.1016\/j.csbj.2025.07.038","article-title":"Tokenization and deep learning architectures in genomics: A comprehensive review","volume":"27","year":"2025","journal-title":"Computational and Structural Biotechnology Journal"},{"key":"2026092813251460000_2025.12.12.693901v4.22","doi-asserted-by":"crossref","first-page":"992","DOI":"10.1016\/j.csbj.2025.03.007","article-title":"Benchmarking DNA large language models on quadruplexes","volume":"27","year":"2025","journal-title":"Computational and Structural Biotechnology Journal"},{"key":"2026092813251460000_2025.12.12.693901v4.23","article-title":"Don\u2019t stop pretraining: Adapt language models to domains and tasks","year":"2020","journal-title":"arXiv preprint"},{"key":"2026092813251460000_2025.12.12.693901v4.24","unstructured":"National Center for Biotechnology Information. NCBI Virus. https:\/\/www.ncbi.nlm.nih.gov\/labs\/virus\/vssi\/#\/. Accessed: November 2024."},{"key":"2026092813251460000_2025.12.12.693901v4.25","doi-asserted-by":"publisher","DOI":"10.1093\/NAR\/GKAC1003"},{"key":"2026092813251460000_2025.12.12.693901v4.26","doi-asserted-by":"publisher","DOI":"10.46234\/ccdcw2021.255"},{"key":"2026092813251460000_2025.12.12.693901v4.27","unstructured":"Los Alamos National Laboratory. HIV Sequence Database. https:\/\/www.hiv.lanl.gov\/. Accessed: November 2024."},{"key":"2026092813251460000_2025.12.12.693901v4.28","doi-asserted-by":"publisher","DOI":"10.1093\/nar\/gks1022"},{"key":"2026092813251460000_2025.12.12.693901v4.29","unstructured":"Raphael Mourad. Mistral-DNA-v1-138M-hg38. Hugging Face. Online resource."},{"key":"2026092813251460000_2025.12.12.693901v4.30","doi-asserted-by":"crossref","unstructured":"Ketan Rajshekhar Shahapure and Charles Nicholas . Cluster quality analysis using silhouette score. In Proceedings of the 2020 IEEE 7th International Conference on Data Science and Advanced Analytics (DSAA), pages 747\u2013748, Sydney, Australia, 2020.","DOI":"10.1109\/DSAA49011.2020.00096"},{"key":"2026092813251460000_2025.12.12.693901v4.31","doi-asserted-by":"publisher","DOI":"10.1109\/TPAMI.1979.4766909"},{"key":"2026092813251460000_2025.12.12.693901v4.32","doi-asserted-by":"publisher","DOI":"10.1109\/TPAMI.2002.1114856"},{"key":"2026092813251460000_2025.12.12.693901v4.33","article-title":"Decoupled weight decay regularization","year":"2019","journal-title":"arXiv preprint"},{"key":"2026092813251460000_2025.12.12.693901v4.34","doi-asserted-by":"publisher","DOI":"10.1093\/bioinformatics\/bty630"},{"key":"2026092813251460000_2025.12.12.693901v4.35","doi-asserted-by":"publisher","DOI":"10.2307\/3001968"},{"key":"2026092813251460000_2025.12.12.693901v4.36","doi-asserted-by":"crossref","unstructured":"Haonan Feng et al. Benchmarking DNA foundation models for genomic sequence classification. bioRxiv, page 2024.08.16.608288, 2024.","DOI":"10.1038\/s41467-025-65823-8"},{"key":"2026092813251460000_2025.12.12.693901v4.37","doi-asserted-by":"crossref","first-page":"133","DOI":"10.1016\/j.meegid.2006.04.002","article-title":"Virus species and virus identification: Past and current controversies","volume":"7","year":"2007","journal-title":"Infection, Genetics and Evolution"},{"key":"2026092813251460000_2025.12.12.693901v4.38","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pone.0006949"},{"key":"2026092813251460000_2025.12.12.693901v4.39","first-page":"4765","article-title":"A unified approach to interpreting model predictions","volume":"30","year":"2017","journal-title":"In Advances in Neural Information Processing Systems"},{"key":"2026092813251460000_2025.12.12.693901v4.40","doi-asserted-by":"publisher","DOI":"10.1126\/science.ado9336"},{"key":"2026092813251460000_2025.12.12.693901v4.41","unstructured":"Michael Poli , Stefano Massaroli , Eric Nguyen , Daniel Y. Fu , Tri Dao , Stephen Baccus , Yoshua Bengio , Stefano Ermon , and Christopher R\u00e9. Hyena hierarchy: Towards larger convolutional language models. In Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Research, pages 28043\u201328078. PMLR, 2023."},{"key":"2026092813251460000_2025.12.12.693901v4.42","doi-asserted-by":"publisher","DOI":"10.1128\/mBio.02248-18"},{"key":"2026092813251460000_2025.12.12.693901v4.43","doi-asserted-by":"publisher","DOI":"10.1126\/science.abc4730"},{"key":"2026092813251460000_2025.12.12.693901v4.44","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.2206509119"},{"key":"2026092813251460000_2025.12.12.693901v4.45","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pbio.3001284"},{"key":"2026092813251460000_2025.12.12.693901v4.46","doi-asserted-by":"crossref","unstructured":"Raveen Rathnasinghe , Sonia Jangra , Anastasija Cupic , et al. The N501Y mutation in SARS-CoV-2 spike leads to morbidity in obese and aged mice and is neutralized by convalescent and post-vaccination human sera. medRxiv, page 2021.01.19.21249592, 2021.","DOI":"10.1101\/2021.01.19.21249592"},{"issue":"1","key":"2026092813251460000_2025.12.12.693901v4.47","doi-asserted-by":"crossref","first-page":"14","DOI":"10.1186\/s13062-022-00325-x","article-title":"Long-range enhancement of N501Y-endowed mouse infectivity of SARS-CoV-2 by the non-RBD mutations of Ins215KLRS and H655Y","volume":"17","year":"2022","journal-title":"Biology Direct"},{"issue":"1","key":"2026092813251460000_2025.12.12.693901v4.48","doi-asserted-by":"crossref","first-page":"64","DOI":"10.1007\/s40484-019-0187-4","article-title":"Identifying viruses from metagenomic data using deep learning","volume":"8","year":"2020","journal-title":"Quantitative Biology"},{"issue":"3","key":"2026092813251460000_2025.12.12.693901v4.49","doi-asserted-by":"crossref","first-page":"1840","DOI":"10.1109\/TCBB.2020.3044575","article-title":"RNN-VirSeeker: A deep learning method for identification of short viral sequences from metagenomes","volume":"19","year":"2022","journal-title":"IEEE\/ACM Transactions on Computational Biology and Bioinformatics"}],"container-title":[],"original-title":[],"link":[{"URL":"https:\/\/syndication.highwire.org\/content\/doi\/10.64898\/2025.12.12.693901","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,28]],"date-time":"2026-09-28T20:25:32Z","timestamp":1790627132000},"score":1,"resource":{"primary":{"URL":"http:\/\/biorxiv.org\/lookup\/doi\/10.64898\/2025.12.12.693901"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,12,13]]},"references-count":49,"URL":"https:\/\/doi.org\/10.64898\/2025.12.12.693901","relation":{},"subject":[],"published":{"date-parts":[[2025,12,13]]},"subtype":"preprint"}}