Cell context-specific Synthetic lethality Prediction and Mechanism Analysis

Abstract

AbstractSynthetic lethality (SL) holds significant promise as a targeted cancer therapy by selectively eliminating tumor cells while sparing normal cells. However, the discovery of SL gene pairs has encountered tremendous challenges, including high costs and low efficiency of experimental methods. Current computational approaches only provide limited insights because of overlooking the crucial aspects of cellular context dependency and mechanistic understanding of SL pairs. To overcome these challenges, we have developed SLWise, a deep-learning model capable of predicting SL interactions in diverse cellular backgrounds. We evaluated SLWise using a real world ground truth standard. The evaluation demonstrated that SLWise outperformed benchmark models in SL prediction. Additionally, we proposed a novel analysis scheme called SLAD-CE (SyntheticLethalAssociated GeneDetection andCell DamageEvaluation) for the identification of abnormal essential genes induced by SL gene pairs and tracking the extent of cell damage. Leveraging the cell-line-specific input feature L1000 and employing Gene Set Enrichment Analysis (GSEA), SLAD-CE provides valuable insights into the underlying mechanisms of SLWise-predicted gene pairs. The combined utilization of SLWise and SLAD-CE offers an approach for predicting and analyzing SL interactions in specific cellular contexts. Our findings highlight the potential of SLWise and SLAD-CE in advancing SL-based therapies by improving prediction accuracy and enhancing mechanistic understanding, ultimately contributing to the development of effective precision treatments for cancer.