The paper introduces SIDISH, an innovative deep learning framework designed to bridge the gap between high-resolution single-cell RNA sequencing and the extensive clinical reach of bulk RNA sequencing. By utilizing an iterative learning process involving variational autoencoders and deep Cox regression, the system identifies specific high-risk cell subpopulations and genetic biomarkers directly linked to poor patient survival. Research findings demonstrate its versatility across various diseases, including pancreatic, breast, and lung cancers, where it successfully maps cellular dynamics to clinical phenotypes. The framework also extends to spatial transcriptomics, allowing researchers to pinpoint high-risk cells within their native tissue architecture. Furthermore, SIDISH features an in silico perturbation module that simulates genetic interventions to prioritize potential therapeutic targets for precision medicine. Benchmarking evaluations show that SIDISH outperforms existing computational tools, offering a more robust and scalable approach for biomarker discovery and personalized treatment strategies.
References:
Jolasun Y, Song K, Zheng Y, et al. SIDISH integrates single-cell and bulk transcriptomics to identify high-risk cells and guide precision therapeutics through in silico perturbation[J]. Nature Communications, 2025.