Sach Mukherjee

This study evaluates genomic foundation models' ability to improve common variant discovery in GWAS and finds they do not outperform standard functional annotations.

This paper introduces a method that uses causal signals from genetic screens to improve the robustness and interpretability of AI models for cancer medicine, showing effective generalization across data and biological contexts.

CopyClust improves breast cancer subtype classification using DNA copy number data alone, achieving higher accuracy than existing methods.

This paper integrates transcriptome and chromatin state data to develop a scalable pipeline for drug discovery, predicting cancer cell line sensitivity and guiding compound development.