Synthetic lethal connectivity and graph transformer improve synthetic lethality prediction.

Area of Science:

• Computational Biology and Bioinformatics
• Genomics and Cancer Research

Background:

• Synthetic lethality (SL) is a promising strategy for identifying novel cancer targets.
• Current CRISPR double-knockout (CDKO) technologies are limited in genome-wide screening capacity.
• Scalable and generalizable SL properties are lacking, hindering predictive modeling efforts.

Purpose of the Study:

• To develop a robust, scalable, and generalizable SL prediction model.
• To improve the selection of gene pairs for CDKO experiments.
• To identify novel SL interactions for cancer therapy.

Main Methods:

• Developed a novel two-step multilayer encoder for individual sample-specific SL prediction (MLEC-iSL).
• MLEC-iSL integrates gene, graph, and transformer encoders to predict SL connectivity and interactions.
• Validated MLEC-iSL performance using K562 and Jurkat cell lines and a CDKO experiment in 22Rv1 cells.

Main Results:

• MLEC-iSL achieved high prediction performance in K562 (AUPR 0.73, AUC 0.72) and Jurkat (AUPR 0.73, AUC 0.71) cells, outperforming existing methods.
• Experimental validation in 22Rv1 cells showed a 46.8% SL rate among 987 selected gene pairs.
• The study revealed SL dependencies between apoptosis and mitosis cell death pathways.

Conclusions:

• SL connectivity is a scalable and generalizable property suitable for predictive modeling.
• MLEC-iSL demonstrates superior performance in predicting SL interactions and guiding CDKO experiments.
• The findings highlight potential therapeutic targets and pathways in cancer through synthetic lethality.