biostudies-literatureUnknownHaber ENIBIB NIH HHSNIDA NIH HHSNHGRI NIH HHSNCI NIH HHSNIH HHSFoundation 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> _G), an expression encoder (<i>F</i> _E), and a "cell sentence" constructor (<i>F</i> _C) with submodules (order, sequence, and reduce) enabling fine-grained control and attribution. Using a transformer trained from scratch, we evalubioRxiv : the preprint server for biology2025.11.09.687403https://www.ebi.ac.uk/biostudies/studies/S-EPMC12642435biostudies-literatureHeimdall: A Modular Framework for Tokenization in Single-Cell Foundation Models.PMC12642435R21 DA061481R03 OD039980UM1 HG011593R01 HG012303UH3 CA268202T32 EB009403R01 HG007352Haber EAlam SLiang SKrieger STrop EHo NYang MLiu RMa JfalseHeimdall: A Modular Framework for Tokenization in Single-Cell Foundation Models.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> _G), an expression encoder (<i>F</i> _E), and a "cell sentence" constructor (<i>F</i> _C) with submodules (order, sequence, and reduce) enabling fine-grained control and attribution. Using a transformer trained from scratch, we evalu2025-01-01T00:00:00Z2025 Nov2026-06-10T03:13:54.034Z2026-06-10T03:07:46.916ZS-EPMC126424354129291310.1101/2025.11.09.687403