Ning Shen

This study benchmarks eight pre-trained genomic language models on RNA sequence prediction tasks, finding that larger models are not always better and that integrating biological information improves performance.

Shusi predicts context-specific protein-protein interactions in cancer using a large language model-enhanced variational graph auto-encoder, identifying actionable vulnerabilities in acute myeloid leukemia.

scFusion detects gene fusions in single-cell RNA sequencing data with high sensitivity and low false discovery rate, successfully identifying fusions missed by bulk methods.

RESA identifies expressed somatic mutations from scRNA-seq data with high precision, outperforming previous methods and revealing insights into intratumor heterogeneity.