A single-cell multimodal foundation model enables cross modality analysis between transcriptome and chromatin

MUTUAL learns a shared RNA–ATAC latent space from 3.0 million paired cells across 400 multiome datasets, enabling more accurate cross-modality matching without refitting.

• Why it matters: Single-cell multiome links RNA and chromatin, but most studies measure only one modality, and prior approaches often require fitting separate models per dataset.
• What they did: MUTUAL keeps scFoundation and EpiAgent frozen, combines their embeddings with linear modality summaries, and trains on paired cells to align RNA and ATAC in one space.
• The result: Without refitting on ten held-out datasets, it matched RNA and ATAC profiles more accurately than donor-fitted methods, improved ATAC-only annotation, and transferred transcription-factor motif activity to RNA-only cells.

AI summary of the abstract · may contain errors · not medical advice · read the paper.