Learning gene co-regulation across cells with Multicellular Transformer
Understanding how genes coordinate across cells is central to explaining tissue function and disease, yet most existing models still take a single cell as input and therefore emphasize cell-intrinsic regulation. We present the Multicellular Transformer (MCT), a foundation model built around a different assumption: cells function together within tissues, so the appropriate unit of input is not one cell in isolation but a set of cells from the same sample. By jointly modeling multiple cells, MCT learns context-aware gene embeddings that capture gene-gene interactions across the tissue environment. Trained on a large set of tissue samples (>4200) across more than 30 tissue types, MCT recovers ligand-receptor interactions, infers cross-cell signaling programs, and reveals spatial co-localization patterns directly from scRNA-seq data. Its contextual embeddings identify disease-specific rewiring of gene neighborhoods in systemic lupus erythematosus and COVID-19 and predict population-level perturbation effects in cancer without additional training. By linking gene-level regulation to tissue-scale organization, MCT provides a scalable framework for modeling multicellular biology and enables in silico discovery of therapeutic targets and tissue-specific dependencies.
Authors
- Tommaso Biancalani (ORCID: https://orcid.org/0000-0001-9104-9755)
- Maciej Wiatrak
- Kam Hon Hoi (ORCID: https://orcid.org/0000-0003-4158-0446)
- Alma Andersson (ORCID: https://orcid.org/0000-0002-4773-9975)
- Aïcha BenTaieb (ORCID: https://orcid.org/0000-0002-7899-3327)
Institutions
- Genentech (US)
Publication Details
- Journal
- bioRxiv (Cold Spring Harbor Laboratory)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.64898/2026.11.14.766251
- Primary Topic
- Single-cell and spatial transcriptomics
- Type
- preprint