{"database":"biostudies-literature","file_versions":[],"scores":null,"additional":{"omics_type":["Unknown"],"submitter":["Haber E"],"funding":["NIBIB NIH HHS","NIDA NIH HHS","NHGRI NIH HHS","NCI NIH HHS","NIH HHS"],"pubmed_abstract":["Foundation models trained on single-cell RNA-sequencing (scRNA-seq) data have rapidly become powerful tools for single-cell analysis. Their performance, however, depends critically on how cells are tokenized into model inputs - a design space that remains poorly understood. Here, we present Heimdall, a comprehensive framework and open-source toolkit for systematically evaluating tokenization strategies in single-cell foundation models (scFMs). Heimdall decomposes each scFM into modular components: a gene identity encoder (<i>F</i> <sub>G</sub>), an expression encoder (<i>F</i> <sub>E</sub>), and a \"cell sentence\" constructor (<i>F</i> <sub>C</sub>) with submodules (order, sequence, and reduce) enabling fine-grained control and attribution. Using a transformer trained from scratch, we evalu"],"journal":["bioRxiv : the preprint server for biology"],"pagination":["2025.11.09.687403"],"full_dataset_link":["https://www.ebi.ac.uk/biostudies/studies/S-EPMC12642435"],"repository":["biostudies-literature"],"pubmed_title":["Heimdall: A Modular Framework for Tokenization in Single-Cell Foundation Models."],"pmcid":["PMC12642435"],"funding_grant_id":["R21 DA061481","R03 OD039980","UM1 HG011593","R01 HG012303","UH3 CA268202","T32 EB009403","R01 HG007352"],"pubmed_authors":["Haber E","Alam S","Liang S","Krieger S","Trop E","Ho N","Yang M","Liu R","Ma J"],"additional_accession":[]},"is_claimable":false,"name":"Heimdall: A Modular Framework for Tokenization in Single-Cell Foundation Models.","description":"Foundation models trained on single-cell RNA-sequencing (scRNA-seq) data have rapidly become powerful tools for single-cell analysis. Their performance, however, depends critically on how cells are tokenized into model inputs - a design space that remains poorly understood. Here, we present Heimdall, a comprehensive framework and open-source toolkit for systematically evaluating tokenization strategies in single-cell foundation models (scFMs). Heimdall decomposes each scFM into modular components: a gene identity encoder (<i>F</i> <sub>G</sub>), an expression encoder (<i>F</i> <sub>E</sub>), and a \"cell sentence\" constructor (<i>F</i> <sub>C</sub>) with submodules (order, sequence, and reduce) enabling fine-grained control and attribution. Using a transformer trained from scratch, we evalu","dates":{"release":"2025-01-01T00:00:00Z","publication":"2025 Nov","modification":"2026-06-10T03:13:54.034Z","creation":"2026-06-10T03:07:46.916Z"},"accession":"S-EPMC12642435","cross_references":{"pubmed":["41292913"],"doi":["10.1101/2025.11.09.687403"]}}