Short talk

Overview

Short talk at the Data Science for Planetary and Human Health symposium hosted by the Novo Nordisk Foundation Science Cluster (Copenhagen, 10–12 June 2024)

• Symposium: Data Science for Planetary and Human Health, Novo Nordisk Foundation Science Cluster
• Session: Session 7 (Tuesday 11 June 2024), chaired by Richard Dennis
• Format: short talk

Abstract

Exploiting synthetic lethal interactions is a promising approach for the development of new targeted cancer therapies. Paralogs, genes that arise from gene duplication, are enriched in synthetic lethal interactions, making them a valuable source of therapeutic targets. However, to date, only a minority of paralog pairs have been experimentally tested for synthetic lethality, and many pairs have been identified as synthetic lethal only in specific contexts. Consequently, there is a need for computational tools to prioritize those paralog pairs most likely to be synthetic lethal and to identify the contexts in which these synthetic lethal interactions will operate. To address these challenges, we collated large-scale CRISPR screens of paralog pairs and annotated them with context-specific information (including expression, mutation, and gene essentiality). We found that single gene essentiality and gene expression are among the features that are most predictive of context-specific synthetic lethality between paralog pairs. Combining multiple features, we developed a random forest classifier that distinguishes SL pairs from non-SL pairs with a prediction performance, ROC AUC of ~0.92. Our classifier outperforms individual features and existing classifiers that ignore context and may help to identify those synthetic lethal pairs that are most likely to operate in specific cancer types of interest. Overall, our study highlights the potential of paralog-based synthetic lethality as a broadly applicable approach for targeting gene loss in cancer and sheds light on features that make paralog pairs likely to be synthetic lethal in specific contexts.