Comprehensive prediction of robust synthetic lethality between paralog pairs in cancer cell lines

Area of Science:

• Genomics
• Cancer Biology
• Computational Biology

Background:

• Paralogs, genes with shared ancestry, can exhibit synthetic lethality, a concept exploitable for cancer treatment.
• Current understanding of synthetic lethal interactions between human paralogs is limited, with only a fraction of pairs investigated.
• Targeting gene loss in cancer through synthetic lethality is a potentially broad therapeutic approach.

Purpose of the Study:

• To identify features predicting synthetic lethality between paralog pairs.
• To develop a machine learning model for predicting and explaining paralog-pair synthetic lethality.
• To assess the predictive accuracy and specificity of the developed model.

Main Methods:

• Analysis of genome-wide CRISPR screens and molecular profiles from over 700 cancer cell lines.
• Identification of predictive features for synthetic lethality, including shared protein-protein interactions and evolutionary conservation.
• Development and validation of a machine learning classifier to predict synthetic lethal paralog pairs.

Main Results:

• Identified key features predictive of synthetic lethality between paralogs.
• Developed a machine learning classifier that accurately predicts synthetic lethality in combinatorial CRISPR screens.
• The classifier can differentiate between universally synthetic lethal pairs and those specific to certain cell lines.

Conclusions:

• Machine learning models can effectively predict synthetic lethality between gene paralogs.
• This approach enhances the potential for broad application of synthetic lethality in cancer therapy.
• Predictive models can guide the selection of paralog pairs for targeted cancer treatments.