Unifying representations of brain cells across modalities and contexts.

A multimodal transformer-based model for single-cell RNA and ATAC data.

From a cell to its representation

Experimental measurements Feature sampling Model training Cell Gene Peak View I View II Encoder Attention heads QKV seqposstd Perceiver Latents Q crossattention selfattention ×N View I View II Cellrepresentation Task heads Reconstruction View I View II + VISReg View I View II
0/3

Consistent across context sizes

Aligned across modalities