Learning epistatic gene interactions from perturbation screens.

Area of Science:

• Genomics
• Systems Biology
• Computational Biology

Background:

• Complex diseases require combinatorial therapies targeting multiple genes.
• Identifying epistatic gene interactions is crucial but experimentally challenging due to combinatorial complexity.
• RNA interference (RNAi) screens offer a platform for studying gene interactions, leveraging inherent off-target effects.

Purpose of the Study:

• To present a computational model for inferring pairwise epistatic gene interactions, including synthetic lethality, from siRNA-based screens.
• To evaluate the performance of regression-based methods (glinternet and xyz) for estimating epistasis in high-dimensional genetic data.
• To demonstrate the model's applicability in identifying therapeutic targets from existing RNAi screening data.

Main Methods:

• Developed a regression-based model to estimate conditional and marginal epistasis from perturbation data.
• Compared glinternet and xyz for feature selection in high-dimensional settings.
• Simulated data from RNAi screening libraries to assess model accuracy under various parameters.
• Applied the model to published siRNA perturbation screens to identify epistatic interactions.

Main Results:

• The model successfully estimates epistatic interactions, including synthetic lethal pairs.
• glinternet demonstrated high accuracy in identifying epistatic gene pairs from simulated RNAi data, especially in high dimensions.
• xyz performed well on lower-dimensional datasets but was less accurate with more genes.
• The model identified both known and novel epistatic interactions in pathogen-related siRNA screens.

Conclusions:

• The developed model provides an efficient method for discovering epistatic gene interactions from large-scale RNAi screening data.
• The findings support the use of combinatorial perturbation screens and computational inference for identifying therapeutic targets.
• The model is broadly applicable to existing public RNAi screening datasets for uncovering gene interactions relevant to complex diseases.