Uncovering cancer vulnerabilities by machine learning prediction of synthetic lethality

Area of Science:

• Computational Biology
• Genomics
• Cancer Research

Background:

• Synthetic lethality (SL) is a genetic interaction where combined perturbations cause cell death, offering a strategy for tumor-specific targeting.
• CRISPR viability screens are crucial for identifying cancer vulnerabilities, but systematic inference of genetic interactions remains a challenge.

Purpose of the Study:

• To introduce PAn-canceR Inferred Synthetic lethalities (PARIS), a novel machine learning approach for identifying cancer vulnerabilities.
• To systematically infer synthetic lethal interactions using CRISPR screens combined with multi-omics data.

Main Methods:

• PARIS integrates CRISPR viability screening data with genomics and transcriptomics data from hundreds of cancer cell lines in the Cancer Dependency Map.
• The machine learning model predicts synthetic lethal interactions based on integrated multi-omics and screening data.

Main Results:

• PARIS predicted 15 high-confidence synthetic lethal interactions among 549 DNA damage repair (DDR) genes.
• Experimental validation confirmed SL interactions involving CDKN2A, thymidine phosphorylase (TYMP), and thymidylate synthase (TYMS), suggesting patient stratification for TYMS inhibitors.
• A genome-wide mapping identified a synthetic lethal interaction between aldehyde dehydrogenase ALDH2 and BRIP1, highlighting BRIP1 as a potential therapeutic target in ~30% of tumors with low ALDH2 expression.

Conclusions:

• PARIS provides an unbiased, scalable, and adaptable platform for identifying synthetic lethal interactions.
• This approach is expected to enhance cancer therapy by leveraging the increasing availability of cancer genomics data.