{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,9]],"date-time":"2026-06-09T20:18:39Z","timestamp":1781036319974,"version":"3.54.1"},"reference-count":24,"publisher":"Springer Science and Business Media LLC","issue":"3","license":[{"start":{"date-parts":[[2026,1,30]],"date-time":"2026-01-30T00:00:00Z","timestamp":1769731200000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2026,1,30]],"date-time":"2026-01-30T00:00:00Z","timestamp":1769731200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":["Nat Comput Sci"],"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Accurately annotating cell types is essential for extracting biological insight from single-cell RNA sequencing data. Although cell types are naturally organized into hierarchical ontologies, most computational models do not explicitly incorporate this structure into their training objectives. Here, we introduce a hierarchical cross-entropy loss that aligns model objectives with biological structure. Applied to architectures ranging from linear models to transformers, this simple modification improves out-of-distribution performance by 12\u221215% without added computational cost. Critically, we underscore the need to focus on new data generation that improves the connectivity among annotated cell types. Our work suggests that this is likely to yield more generalizable algorithms than would solely increasing model complexity.<\/jats:p>","DOI":"10.1038\/s43588-025-00945-z","type":"journal-article","created":{"date-parts":[[2026,1,30]],"date-time":"2026-01-30T10:02:57Z","timestamp":1769767377000},"page":"243-249","update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":1,"title":["Improving atlas-scale single-cell annotation models with hierarchical cross-entropy loss"],"prefix":"10.1038","volume":"6","author":[{"ORCID":"https:\/\/orcid.org\/0000-0001-6249-0832","authenticated-orcid":false,"given":"Sebastiano","family":"Cultrera di Montesano","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-0607-8488","authenticated-orcid":false,"given":"Davide","family":"D\u2019Ascenzo","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Srivatsan","family":"Raghavan","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-8601-6040","authenticated-orcid":false,"given":"Ava P.","family":"Amini","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6557-3219","authenticated-orcid":false,"given":"Peter S.","family":"Winter","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0178-8242","authenticated-orcid":false,"given":"Lorin","family":"Crawford","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"297","published-online":{"date-parts":[[2026,1,30]]},"reference":[{"key":"945_CR1","doi-asserted-by":"publisher","first-page":"58","DOI":"10.1126\/science.aan6828","volume":"358","author":"MJT Stubbington","year":"2017","unstructured":"Stubbington, M. J. T., Rozenblatt-Rosen, O., Regev, A. & Teichmann, S. A. Single-cell transcriptomics to explore the immune system in health and disease. Science 358, 58\u201363 (2017).","journal-title":"Science"},{"key":"945_CR2","doi-asserted-by":"publisher","DOI":"10.7554\/eLife.27041","volume":"6","author":"A Regev","year":"2017","unstructured":"Regev, A. et al. Science forum: the Human Cell Atlas. eLife 6, e27041 (2017).","journal-title":"eLife"},{"key":"945_CR3","first-page":"D886","volume":"53","author":"S Abdulla","year":"2024","unstructured":"Abdulla, S. et al. CZ CELLxGENE Discover: a single-cell data platform for scalable exploration, analysis and modeling of aggregated data. Nucleic Acids Res. 53, D886\u2013D900 (2024).","journal-title":"Nucleic Acids Res."},{"key":"945_CR4","doi-asserted-by":"publisher","first-page":"194","DOI":"10.1186\/s13059-019-1795-z","volume":"20","author":"T Abdelaal","year":"2019","unstructured":"Abdelaal, T. et al. A comparison of automatic cell identification methods for single-cell RNA sequencing data. Genome Biol. 20, 194 (2019).","journal-title":"Genome Biol."},{"key":"945_CR5","doi-asserted-by":"publisher","first-page":"961","DOI":"10.1016\/j.csbj.2021.01.015","volume":"19","author":"G Pasquini","year":"2021","unstructured":"Pasquini, G., Arias, J. E. R., Sch\u00e4fer, P. & Busskamp, V. Automated methods for cell type annotation on scRNA-seq data. Comput. Struct. Biotechnol. J. 19, 961\u2013969 (2021).","journal-title":"Comput. Struct. Biotechnol. J."},{"key":"945_CR6","doi-asserted-by":"publisher","DOI":"10.1126\/science.abl5197","volume":"376","author":"CD Conde","year":"2022","unstructured":"Conde, C. D. et al. Cross-tissue immune cell analysis reveals tissue-specific features in humans. Science 376, eabl5197 (2022).","journal-title":"Science"},{"key":"945_CR7","doi-asserted-by":"publisher","first-page":"2731","DOI":"10.1038\/s41588-024-01993-3","volume":"56","author":"C Ergen","year":"2024","unstructured":"Ergen, C. et al. Consensus prediction of cell type labels in single-cell data with popV. Nat. Genet. 56, 2731\u20132738 (2024).","journal-title":"Nat. Genet."},{"key":"945_CR8","doi-asserted-by":"publisher","first-page":"bbab035","DOI":"10.1093\/bib\/bbab035","volume":"22","author":"Y Huang","year":"2021","unstructured":"Huang, Y. & Zhang, P. Evaluation of machine learning approaches for cell-type identification from single-cell transcriptomics data. Brief. Bioinform. 22, bbab035 (2021).","journal-title":"Brief. Bioinform."},{"key":"945_CR9","doi-asserted-by":"publisher","first-page":"1129","DOI":"10.1038\/s41556-021-00787-7","volume":"23","author":"D Osumi-Sutherland","year":"2021","unstructured":"Osumi-Sutherland, D. et al. Cell type ontologies of the Human Cell Atlas. Nat. Cell Biol. 23, 1129\u20131135 (2021).","journal-title":"Nat. Cell Biol."},{"key":"945_CR10","unstructured":"Jupp, S., Burdett, T., Leroy, C. & Parkinson, H. E. A new ontology lookup service at EMBL-EBI. In Proc. 8th International Conference on Semantic Web Applications and Tools for Life Sciences Vol. 1546 (eds Malone, J. et al.) 118\u2013119 (CEUR-WS.org, 2015)."},{"key":"945_CR11","doi-asserted-by":"publisher","first-page":"6611","DOI":"10.1038\/s41467-024-51059-5","volume":"15","author":"F Fischer","year":"2024","unstructured":"Fischer, F. et al. scTab: scaling cross-tissue single-cell annotation models. Nat. Commun. 15, 6611 (2024).","journal-title":"Nat. Commun."},{"key":"945_CR12","doi-asserted-by":"publisher","first-page":"5556","DOI":"10.1038\/s41467-021-25725-x","volume":"12","author":"S Wang","year":"2021","unstructured":"Wang, S. et al. Leveraging the cell ontology to classify unseen cell types. Nat. Commun. 12, 5556 (2021).","journal-title":"Nat. Commun."},{"key":"945_CR13","doi-asserted-by":"publisher","first-page":"101913","DOI":"10.1016\/j.isci.2020.101913","volume":"24","author":"MN Bernstein","year":"2019","unstructured":"Bernstein, M. N., Ma, Z., Gleicher, M. & Dewey, C. N. CellO: comprehensive and hierarchical cell type classification of human cells with the Cell Ontology. iScience 24, 101913 (2019).","journal-title":"iScience"},{"key":"945_CR14","doi-asserted-by":"publisher","first-page":"1053","DOI":"10.1038\/s41592-018-0229-2","volume":"15","author":"R Lopez","year":"2018","unstructured":"Lopez, R., Regier, J., Cole, M., Jordan, M. I. & Yosef, N. Deep generative modeling for single-cell transcriptomics. Nat. Methods 15, 1053\u20131058 (2018).","journal-title":"Nat. Methods"},{"key":"945_CR15","doi-asserted-by":"publisher","first-page":"852","DOI":"10.1038\/s42256-022-00534-z","volume":"4","author":"W Wang","year":"2022","unstructured":"Wang, W. et al. scBERT as a large-scale pretrained deep language model for cell type annotation of single-cell RNA-seq data. Nat. Mach. Intell. 4, 852\u2013866 (2022).","journal-title":"Nat. Mach. Intell."},{"key":"945_CR16","doi-asserted-by":"publisher","first-page":"223","DOI":"10.1038\/s41467-023-35923-4","volume":"14","author":"J Chen","year":"2023","unstructured":"Chen, J. et al. Transformer for one stop interpretable cell type annotation. Nat. Commun. 14, 223 (2023).","journal-title":"Nat. Commun."},{"key":"945_CR17","doi-asserted-by":"publisher","first-page":"bbad195","DOI":"10.1093\/bib\/bbad195","volume":"24","author":"J Xu","year":"2023","unstructured":"Xu, J., Zhang, A., Liu, F., Chen, L. & Zhang, X. CIForm as a Transformer-based model for cell-type annotation of large-scale single-cell RNA-seq data. Brief. Bioinform. 24, bbad195 (2023).","journal-title":"Brief. Bioinform."},{"key":"945_CR18","doi-asserted-by":"crossref","unstructured":"Arik, S. & Pfister, T. TabNet: attentive interpretable tabular learning. In Proc. AAAI Conference on Artificial Intelligence Vol. 35 (eds Leyton-Brown, K. & Mausam) 6679\u20136687 (AAAI, 2021).","DOI":"10.1609\/aaai.v35i8.16826"},{"key":"945_CR19","doi-asserted-by":"publisher","first-page":"1085","DOI":"10.1038\/s41586-024-08411-y","volume":"638","author":"G Heimberg","year":"2024","unstructured":"Heimberg, G. et al. A cell atlas foundation model for scalable search of similar human cells. Nature 638, 1085\u20131094 (2024).","journal-title":"Nature"},{"key":"945_CR20","doi-asserted-by":"publisher","first-page":"616","DOI":"10.1038\/s41586-023-06139-9","volume":"618","author":"CV Theodoris","year":"2023","unstructured":"Theodoris, C. V. et al. Transfer learning enables predictions in network biology. Nature 618, 616\u2013624 (2023).","journal-title":"Nature"},{"key":"945_CR21","doi-asserted-by":"publisher","first-page":"1470","DOI":"10.1038\/s41592-024-02201-0","volume":"21","author":"H Cui","year":"2024","unstructured":"Cui, H. et al. scGPT: toward building a foundation model for single-cell multi-omics using generative AI. Nat. Methods 21, 1470\u20131480 (2024).","journal-title":"Nat. Methods"},{"key":"945_CR22","doi-asserted-by":"publisher","first-page":"101","DOI":"10.1186\/s13059-025-03574-x","volume":"26","author":"KZ Kedzierska","year":"2025","unstructured":"Kedzierska, K. Z., Crawford, L., Amini, A. P. & Lu, A. X. Zero-shot evaluation reveals limitations of single-cell foundation models. Genome Biol. 26, 101 (2025).","journal-title":"Genome Biol."},{"key":"945_CR23","doi-asserted-by":"publisher","unstructured":"Cultrera di Montesano, S. et al. Hierarchical cross-entropy loss improves atlas-scale single-cell annotation models. Zenodo https:\/\/doi.org\/10.5281\/zenodo.17211022 (2025).","DOI":"10.5281\/zenodo.17211022"},{"key":"945_CR24","doi-asserted-by":"publisher","unstructured":"D\u2019Ascenzo, D. microsoft\/hce-classification. zenodo https:\/\/doi.org\/10.5281\/zenodo.17554556 (2025).","DOI":"10.5281\/zenodo.17554556"}],"container-title":["Nature Computational Science"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.nature.com\/articles\/s43588-025-00945-z.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s43588-025-00945-z","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s43588-025-00945-z.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,26]],"date-time":"2026-03-26T23:02:12Z","timestamp":1774566132000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.nature.com\/articles\/s43588-025-00945-z"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,1,30]]},"references-count":24,"journal-issue":{"issue":"3","published-online":{"date-parts":[[2026,3]]}},"alternative-id":["945"],"URL":"https:\/\/doi.org\/10.1038\/s43588-025-00945-z","relation":{},"ISSN":["2662-8457"],"issn-type":[{"value":"2662-8457","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,1,30]]},"assertion":[{"value":"28 May 2025","order":1,"name":"received","label":"Received","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"12 December 2025","order":2,"name":"accepted","label":"Accepted","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"30 January 2026","order":3,"name":"first_online","label":"First Online","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"6 February 2026","order":5,"name":"change_date","label":"Change Date","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"Update","order":6,"name":"change_type","label":"Change Type","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"In the version of this article initially published, refs. <sup>23<\/sup> (","order":7,"name":"change_details","label":"Change Details","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"https:\/\/doi.org\/10.5281\/zenodo.17211022","URL":"https:\/\/doi.org\/10.5281\/zenodo.17211022","order":8,"name":"change_details","label":"Change Details","group":{"name":"ArticleHistory","label":"Article History"}},{"value":") and <sup>24<\/sup> (","order":9,"name":"change_details","label":"Change Details","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"https:\/\/doi.org\/10.5281\/zenodo.17554556","URL":"https:\/\/doi.org\/10.5281\/zenodo.17554556","order":10,"name":"change_details","label":"Change Details","group":{"name":"ArticleHistory","label":"Article History"}},{"value":") were not listed correctly with doi format, as is now amended in the HTML and PDF versions of the article.","order":11,"name":"change_details","label":"Change Details","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"S.R. holds equity in Amgen. S.R. and P.S.W. receive research funding from Microsoft. A.P.A. and L.C. are employees of Microsoft and own equity in Microsoft. S.C.d.M. reports compensation from Engine Ventures and Mobius Biotechnology GmbH for consulting\/speaking unrelated to this work. P.S.W. reports compensation from Engine Ventures and AbbVie for consulting\/speaking unrelated to this work. The other authors have no competing interests.","order":1,"name":"Ethics","group":{"name":"EthicsHeading","label":"Competing interests"}}]}}