Matrix (factorization) reloaded: flexible methods for imputing genetic interactions with cross-species and side...

Area of Science:

• Computational biology
• Systems biology
• Genetics

Background:

• Genetic interactions (GIs) provide insights into cellular functions and have translational potential.
• High-throughput experimental systems for measuring GIs have advanced, alongside computational methods for inferring unknown interactions.
• Existing computational methods are largely limited to baker's yeast, require specific side information, are computationally expensive, and struggle with scarce data.

Purpose of the Study:

• To address limitations in current genetic interaction imputation methods.
• To present a new, flexible imputation framework, Extensible Matrix Factorization (EMF).
• To enable exploitation of cross-species and arbitrary side information for improved imputation.

Main Methods:

• Developed and presented the Extensible Matrix Factorization (EMF) framework.
• EMF utilizes composable models to integrate cross-species GI data and arbitrary side information (e.g., protein-protein interaction networks).
• Conducted rigorous experiments on matched GI datasets from baker's and fission yeast, including genome-scale data.

Main Results:

• EMF models incorporating side and cross-species information significantly improve genetic interaction imputation, particularly in data-scarce scenarios.
• EMF outperforms state-of-the-art deep learning methods in imputation accuracy.
• EMF demonstrates substantially lower computational cost compared to existing methods, even when using less data.

Conclusions:

• The EMF framework offers a powerful and efficient approach for imputing genetic interactions.
• EMF's ability to leverage cross-species and side information enhances its applicability and performance, especially in challenging data settings.
• EMF represents a significant advancement in computational methods for genetic interaction analysis.