DISCERN: inferring drug sensitivity from single-cell transcriptomes using cell-type-specific genetic interaction networks.

Abstract

Background: Genetic interactions, including synthetic lethality (SL) and synthetic viability (SV), are crucial for understanding tumor-specific vulnerabilities and mechanisms of drug resistance. However, predicting drug response at single-cell resolution based on SL and SV remains challenging.

Methods: Here, we construct a large-scale atlas of cell-type-specific SL and SV networks across 14 human cancers using scRNA-seq datasets. Based on this atlas, we develop DISCERN (Drug response Inference from Single-Cell gEnetic inteRactioNs), a novel computational framework designed to infer single-cell drug sensitivity by utilizing malignant cell-specific genetic interactions. We also establish CellGIdb, an interactive portal that provides the cell-type-specific genetic interaction networks.

Results: We reconstruct cell-type-specific genetic interaction networks across cancers, revealing both shared and distinct patterns among cell types. Notably, SL and SV interactions derived from malignant cells and T cells exhibit prognostic value and correlate with response to immunotherapy. DISCERN effectively infers tumor cell-specific drug sensitivity in scRNA-seq datasets from lung and breast cancers. DISCERN demonstrates improved predictive performance compared with existing computational methods. CellGIdb provides user-friendly analytical tools to facilitate the exploration of genetic interactions' roles in drug response and immunotherapy.

Conclusions: Collectively, this study provides a comprehensive atlas and a novel computational framework, DISCERN, for interpreting drug responses in the context of genetic interactions at single-cell resolution. The publicly available CellGIdb ( https://biodata.hrbmu.edu.cn/CellGIdb/index.html ) resource will support further exploration of cell-type-specific vulnerabilities in cancer therapy.