OmniCell: Unified Foundation Modeling of Single-Cell and Spatial Transcriptomics for Cellular and Molecular Insights

A cells transcriptional programme is not fully defined by gene expression alone, but by the tissue context in which that programme is enacted. Single-cell RNA sequencing resolves molecular identity after dissociation, whereas spatial transcriptomics preserves tissue architecture but remains constrained by assay-specific sparsity and gene coverage. Here we present OmniCell, a tissue-contextual transcriptomic foundation model pretrained on 67 million dissociated and spatially resolved profiles. By integrating gene identity, expression magnitude and tissue context, OmniCell links transcriptional programmes to the cellular neighbourhoods and anatomical contexts in which they operate. OmniCell organised transcriptomes across molecular, cellular and tissue scales. It recovered cell-type-specific programmes and tissue-aligned gene modules, preserved robust cell-state structure across batches, species and rare populations, and improved the reconstruction of spatial cell identity, anatomical domains and cell-type composition. In human liver cancer Stereo-seq data, OmniCell resolved a tumour-margin transition zone characterised by immune infiltration, acute-phase inflammation, coagulation/complement activity and metallothionein-linked metal-ion detoxification. Contextual gene-embedding similarity analysis showed that gene relationships differed across tumour core, transition-zone and paratumour/adjacent non-malignant niches, indicating that OmniCell captures tissue-dependent gene function rather than expression similarity alone. In mouse brain development and macaque cortex, spatial virtual perturbations mapped regulatory genes onto stage- and region-specific anatomical programmes. Together, these results establish tissue context as a primary axis of transcriptomic representation and provide a framework for studying how cellular programmes acquire context-dependent biological meaning in intact tissues.