Conception

Cisformer is a novel cross-modal deep learning model based on the Transformer architecture. It enables bidirectional prediction and association between cis-regulatory elements and genes at the single-cell level, with high efficiency and accuracy. logo

To meet real-world application needs and improve model performance:

  • For the RNA2ATAC task, a trained Cisformer can generate high-quality scATAC-seq data from scRNA-seq inputs.

  • For the ATAC2RNA task, the model can generate pseudo-scRNA-seq data and construct a highly accurate cis-regulatory interaction matrix between cis-elements and genes.