Wei Jin

DANCE provides a benchmark platform for single-cell analysis with 32 methods across 21 datasets, enabling easy reproduction and development of deep learning models.

Treating cells as tokens and tissues as sentences, with spatial transcriptomics added during pre-training, gives a single-cell foundation model the authors report outperforms existing models and runs far faster.

The paper introduces a large-scale spatial transcriptomic dataset and a Graph Neural Network-based method for cell type deconvolution in human immuno-oncology.