{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,27]],"date-time":"2026-10-27T17:56:11Z","timestamp":1793123771460,"version":"4.5.0"},"reference-count":48,"publisher":"Oxford University Press (OUP)","issue":"5","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2026,5,3]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:sec>\n                    <jats:title>Motivation<\/jats:title>\n                    <jats:p>Large language models (LLMs) have demonstrated strong general reasoning abilities, but applying them to domain-specific tasks such as analysing single-cell RNA sequencing data remains a challenge. A central task in this domain is cell type annotation, which is critical for understanding cellular heterogeneity. Although recent foundation models attempt to automate this process, they typically annotate cells independently, without considering batch-level context or providing explanatory reasoning. To address this limitation, we introduce the CellPuzzles benchmark, which reformulates cell type annotation as a batch-level reasoning task. CellPuzzles spans diverse tissues, diseases, and donor conditions, and requires reasoning across the batch-level cellular context to ensure label uniqueness.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>We find that off-the-shelf LLMs struggle on this task, with the best baseline (OpenAI o1) achieving only 19.0% batch-level accuracy. To fill this gap, we propose Cell-o1, a 7B LLM trained via supervised fine-tuning on distilled reasoning traces, followed by reinforcement learning with batch-level rewards. Cell-o1 achieves state-of-the-art performance, outperforming OpenAI o1 by over 73% and generalizing well across contexts. Further analysis of training dynamics and reasoning behaviors provides insights into batch-level annotation performance and emergent expert-like reasoning.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Availability and Implementation<\/jats:title>\n                    <jats:p>Code and data are available at https:\/\/github.com\/ncbi-nlp\/cell-o1.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.1093\/bioinformatics\/btag208","type":"journal-article","created":{"date-parts":[[2026,5,1]],"date-time":"2026-05-01T12:20:58Z","timestamp":1777638058000},"source":"Crossref","is-referenced-by-count":5,"title":["<tt>\n                      <b>Cell-o1<\/b>\n                    <\/tt>\n                    : training LLMs to solve single-cell reasoning puzzles with reinforcement learning"],"prefix":"10.1093","volume":"42","author":[{"ORCID":"https:\/\/orcid.org\/0000-0001-9538-848X","authenticated-orcid":false,"given":"Yin","family":"Fang","sequence":"first","affiliation":[{"name":"National Institutes of Health Division of Intramural Research, National Library of Medicine, , 8600 Rockville Pike , Bethesda, MD 20894,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-1268-7239","authenticated-orcid":false,"given":"Qiao","family":"Jin","sequence":"additional","affiliation":[{"name":"National Institutes of Health Division of Intramural Research, National Library of Medicine, , 8600 Rockville Pike , Bethesda, MD 20894,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-8049-5298","authenticated-orcid":false,"given":"Guangzhi","family":"Xiong","sequence":"additional","affiliation":[{"name":"University of Virginia Department of Computer Science, , 351 McCormick Road , Charlottesville, VA 22904,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Bowen","family":"Jin","sequence":"additional","affiliation":[{"name":"University of Illinois at Urbana-Champaign Siebel School of Computing and Data Science, , 201 N Goodwin Ave , Urbana, IL 61801,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xianrui","family":"Zhong","sequence":"additional","affiliation":[{"name":"University of Illinois at Urbana-Champaign Siebel School of Computing and Data Science, , 201 N Goodwin Ave , Urbana, IL 61801,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Siru","family":"Ouyang","sequence":"additional","affiliation":[{"name":"University of Illinois at Urbana-Champaign Siebel School of Computing and Data Science, , 201 N Goodwin Ave , Urbana, IL 61801,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yifan","family":"Yang","sequence":"additional","affiliation":[{"name":"National Institutes of Health Division of Intramural Research, National Library of Medicine, , 8600 Rockville Pike , Bethesda, MD 20894,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9723-3246","authenticated-orcid":false,"given":"Aidong","family":"Zhang","sequence":"additional","affiliation":[{"name":"University of Virginia Department of Computer Science, , 351 McCormick Road , Charlottesville, VA 22904,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jiawei","family":"Han","sequence":"additional","affiliation":[{"name":"University of Illinois at Urbana-Champaign Siebel School of Computing and Data Science, , 201 N Goodwin Ave , Urbana, IL 61801,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9998-916X","authenticated-orcid":false,"given":"Zhiyong","family":"Lu","sequence":"additional","affiliation":[{"name":"National Institutes of Health Division of Intramural Research, National Library of Medicine, , 8600 Rockville Pike , Bethesda, MD 20894,","place":["United States"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"286","published-online":{"date-parts":[[2026,5,4]]},"reference":[{"key":"2026052314563655000_btag208-B1","doi-asserted-by":"crossref","first-page":"194","DOI":"10.1186\/s13059-019-1795-z","article-title":"A comparison of automatic cell identification methods for single-cell RNA sequencing data","volume":"20","author":"Abdelaal","year":"2019","journal-title":"Genome Biol"},{"key":"2026052314563655000_btag208-B2","author":"Achiam","year":"2023"},{"key":"2026052314563655000_btag208-B3","doi-asserted-by":"crossref","first-page":"163","DOI":"10.1038\/s41590-018-0276-y","article-title":"Reference-based analysis of lung single-cell sequencing reveals a transitional profibrotic macrophage","volume":"20","author":"Aran","year":"2019","journal-title":"Nat Immunol"},{"key":"2026052314563655000_btag208-B4","doi-asserted-by":"crossref","first-page":"25","DOI":"10.1038\/75556","article-title":"Gene ontology: tool for the unification of biology","volume":"25","author":"Ashburner","year":"2000","journal-title":"Nat Genet"},{"key":"2026052314563655000_btag208-B5","doi-asserted-by":"crossref","first-page":"442","DOI":"10.1016\/j.tig.2020.03.005","article-title":"Opening the black box: interpretable machine learning for geneticists","volume":"36","author":"Azodi","year":"2020","journal-title":"Trends Genet"},{"key":"2026052314563655000_btag208-B6","author":"Bai","year":"2023"},{"key":"2026052314563655000_btag208-B8","first-page":"1877","article-title":"Language models are few-shot learners","volume":"33","author":"Brown","year":"2020","journal-title":"Adv Neural Inf Process Syst"},{"key":"2026052314563655000_btag208-B9","doi-asserted-by":"crossref","first-page":"411","DOI":"10.1038\/nbt.4096","article-title":"Integrating single-cell transcriptomic data across different conditions, technologies, and species","volume":"36","author":"Butler","year":"2018","journal-title":"Nat Biotechnol"},{"key":"2026052314563655000_btag208-B10","author":"Chen"},{"key":"2026052314563655000_btag208-B11","doi-asserted-by":"crossref","first-page":"483","DOI":"10.1038\/s41551-024-01284-6","article-title":"Simple and effective embedding model for single-cell biology built from Chatgpt","volume":"9","author":"Chen","year":"2025","journal-title":"Nat Biomed Eng"},{"key":"2026052314563655000_btag208-B12","author":"Cornacchia","year":"2023"},{"key":"2026052314563655000_btag208-B13","doi-asserted-by":"crossref","first-page":"1470","DOI":"10.1038\/s41592-024-02201-0","article-title":"Scgpt: toward building a foundation model for single-cell multi-omics using generative ai","volume":"21","author":"Cui","year":"2024","journal-title":"Nat Methods"},{"key":"2026052314563655000_btag208-B14","author":"Fang","year":"2025"},{"key":"2026052314563655000_btag208-B15","first-page":"633","author":"Guo","year":"2025"},{"key":"2026052314563655000_btag208-B16","doi-asserted-by":"crossref","first-page":"1481","DOI":"10.1038\/s41592-024-02305-7","article-title":"Large-scale foundation model on single-cell transcriptomics","volume":"21","author":"Hao","year":"2024","journal-title":"Nat Methods"},{"key":"2026052314563655000_btag208-B17","doi-asserted-by":"crossref","first-page":"1462","DOI":"10.1038\/s41592-024-02235-4","article-title":"Assessing gpt-4 for cell type annotation in single-cell rna-seq analysis","volume":"21","author":"Hou","year":"2024","journal-title":"Nat Methods"},{"key":"2026052314563655000_btag208-B18","author":"Jaech","year":"2024"},{"key":"2026052314563655000_btag208-B19","doi-asserted-by":"crossref","first-page":"522","DOI":"10.1016\/j.cels.2021.05.016","article-title":"Machine learning for perturbational single-cell omics","volume":"12","author":"Ji","year":"2021","journal-title":"Cell Syst"},{"key":"2026052314563655000_btag208-B20","author":"Jiang","year":"2025"},{"key":"2026052314563655000_btag208-B21","volume-title":"Proceedings of the 2nd Conference on Language Modeling (COLM)","author":"Jin"},{"key":"2026052314563655000_btag208-B22","doi-asserted-by":"crossref","first-page":"273","DOI":"10.1038\/s41576-018-0088-9","article-title":"Challenges in unsupervised clustering of single-cell rna-seq data","volume":"20","author":"Kiselev","year":"2019","journal-title":"Nat Rev Genet"},{"key":"2026052314563655000_btag208-B23","first-page":"22199","article-title":"Large language models are zero-shot reasoners","volume":"35","author":"Kojima","year":"2022","journal-title":"Adv Neural Inf Process Syst"},{"key":"2026052314563655000_btag208-B24","doi-asserted-by":"crossref","first-page":"31","DOI":"10.1186\/s13059-020-1926-6","article-title":"Eleven grand challenges in single-cell data science","volume":"21","author":"L\u00e4hnemann","year":"2020","journal-title":"Genome Biol"},{"key":"2026052314563655000_btag208-B25","author":"Levine","year":"2024"},{"key":"2026052314563655000_btag208-B26","author":"Liu","year":"2023"},{"key":"2026052314563655000_btag208-B27","doi-asserted-by":"crossref","first-page":"715","DOI":"10.1038\/s41592-019-0494-8","article-title":"Scgen predicts single-cell perturbation responses","volume":"16","author":"Lotfollahi","year":"2019","journal-title":"Nat Methods"},{"key":"2026052314563655000_btag208-B28","doi-asserted-by":"crossref","first-page":"1053","DOI":"10.1162\/coli_a_00523","article-title":"Large language model instruction following: a survey of progresses and challenges","volume":"50","author":"Lou","year":"2024","journal-title":"Comput Linguist"},{"key":"2026052314563655000_btag208-B29","doi-asserted-by":"crossref","first-page":"e8746","DOI":"10.15252\/msb.20188746","article-title":"Current best practices in single-cell rna-seq analysis: a tutorial","volume":"15","author":"Luecken","year":"2019","journal-title":"Mol Syst Biol"},{"key":"2026052314563655000_btag208-B30","author":"Maharana","year":"2022"},{"key":"2026052314563655000_btag208-B31","doi-asserted-by":"crossref","first-page":"100139","DOI":"10.1016\/j.patter.2020.100139","article-title":"Sctenifoldnet: a machine learning workflow for constructing and comparing transcriptome-wide gene regulatory networks from single-cell data","volume":"1","author":"Osorio","year":"2020","journal-title":"Patterns"},{"key":"2026052314563655000_btag208-B32","doi-asserted-by":"crossref","first-page":"961","DOI":"10.1016\/j.csbj.2021.01.015","article-title":"Automated methods for cell type annotation on scrna-seq data","volume":"19","author":"Pasquini","year":"2021","journal-title":"Comput Struct Biotechnol J"},{"key":"2026052314563655000_btag208-B33","doi-asserted-by":"crossref","first-page":"D886","DOI":"10.1093\/nar\/gkae1142","article-title":"Cz cellxgene discover: a single-cell data platform for scalable exploration, analysis and modeling of aggregated data","volume":"53","author":"Program","year":"2025","journal-title":"Nucleic Acids Res"},{"key":"2026052314563655000_btag208-B34","author":"Renze"},{"key":"2026052314563655000_btag208-B35","author":"Schaefer","year":"2024"},{"key":"2026052314563655000_btag208-B36","author":"Schulman","year":"2017"},{"key":"2026052314563655000_btag208-B37","author":"Shao","year":"2024"},{"key":"2026052314563655000_btag208-B38","author":"Shi","year":"2025"},{"key":"2026052314563655000_btag208-B39","doi-asserted-by":"crossref","first-page":"1888","DOI":"10.1016\/j.cell.2019.05.031","article-title":"Comprehensive integration of single-cell data","volume":"177","author":"Stuart","year":"2019","journal-title":"Cell"},{"key":"2026052314563655000_btag208-B40","doi-asserted-by":"crossref","first-page":"616","DOI":"10.1038\/s41586-023-06139-9","article-title":"Transfer learning enables predictions in network biology","volume":"618","author":"Theodoris","year":"2023","journal-title":"Nature"},{"key":"2026052314563655000_btag208-B41","doi-asserted-by":"crossref","first-page":"bbad493","DOI":"10.1093\/bib\/bbad493","article-title":"Opportunities and challenges for chatgpt and large language models in biomedicine and health","volume":"25","author":"Tian","year":"2023","journal-title":"Brief Bioinform"},{"key":"2026052314563655000_btag208-B42","author":"Touvron","year":"2023"},{"key":"2026052314563655000_btag208-B43","doi-asserted-by":"crossref","first-page":"5233","DOI":"10.1038\/s41598-019-41695-z","article-title":"From Louvain to Leiden: guaranteeing well-connected communities","volume":"9","author":"Traag","year":"2019","journal-title":"Sci Rep"},{"key":"2026052314563655000_btag208-B44","doi-asserted-by":"crossref","first-page":"1491","DOI":"10.1101\/gr.190595.115","article-title":"Defining cell types and states with single-cell genomics","volume":"25","author":"Trapnell","year":"2015","journal-title":"Genome Res"},{"key":"2026052314563655000_btag208-B45","doi-asserted-by":"crossref","first-page":"2247","DOI":"10.1038\/s41596-020-0336-2","article-title":"A scalable scenic workflow for single-cell gene regulatory network analysis","volume":"15","author":"Van de Sande","year":"2020","journal-title":"Nat Protoc"},{"key":"2026052314563655000_btag208-B46","author":"Xiong","year":"2025"},{"key":"2026052314563655000_btag208-B47","doi-asserted-by":"crossref","first-page":"852","DOI":"10.1038\/s42256-022-00534-z","article-title":"Scbert as a large-scale pretrained deep language model for cell type annotation of single-cell rna-seq data","volume":"4","author":"Yang","year":"2022","journal-title":"Nat Mach Intell"},{"key":"2026052314563655000_btag208-B48","author":"Zhao","year":"2025"},{"key":"2026052314563655000_btag208-B49","author":"Zhao","year":"2024"}],"container-title":["Bioinformatics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/academic.oup.com\/bioinformatics\/advance-article-pdf\/doi\/10.1093\/bioinformatics\/btag208\/68209518\/btag208.pdf","content-type":"application\/pdf","content-version":"am","intended-application":"syndication"},{"URL":"https:\/\/academic.oup.com\/bioinformatics\/article-pdf\/42\/5\/btag208\/68209518\/btag208.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/academic.oup.com\/bioinformatics\/article-pdf\/42\/5\/btag208\/68209518\/btag208.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,5,23]],"date-time":"2026-05-23T18:56:45Z","timestamp":1779562605000},"score":1,"resource":{"primary":{"URL":"https:\/\/academic.oup.com\/bioinformatics\/article\/doi\/10.1093\/bioinformatics\/btag208\/8667159"}},"subtitle":[],"editor":[{"given":"Virginie","family":"Uhlmann","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"editor"}]}],"short-title":[],"issued":{"date-parts":[[2026,5]]},"references-count":48,"journal-issue":{"issue":"5","published-print":{"date-parts":[[2026,5,3]]}},"URL":"https:\/\/doi.org\/10.1093\/bioinformatics\/btag208","relation":{},"ISSN":["1367-4803","1367-4811"],"issn-type":[{"value":"1367-4803","type":"print"},{"value":"1367-4811","type":"electronic"}],"subject":[],"published-other":{"date-parts":[[2026,5]]},"published":{"date-parts":[[2026,5]]},"article-number":"btag208"}}