Extensible Matrix Factorization

Extensible Matrix Factorization imputes genetic interactions by applying matrix factorization that leverages cross-species data and arbitrary side information (e.g., kernels from protein-protein interaction networks) to improve GI prediction across species.

Key Features:

• Cross-Species Information Utilization: Incorporates genetic interaction (GI) data from multiple species to enable imputation beyond single-species datasets.
• Integration of Arbitrary Side Information: Accepts diverse side information via kernels derived from sources such as protein-protein interaction networks.
• Composable Models: Provides a set of composable models that can be combined or tailored to different analytical needs.
• Efficiency in Data-Scarce Settings: Maintains robust imputation performance when genetic interaction data are limited.
• Computational Efficiency: Reduces computational cost relative to existing methods, including state-of-the-art deep learning approaches.

Scientific Applications:

• Genomics and Systems Biology: Enables imputation of GIs to support analyses of cellular functions and translational research.
• Comparative Evolutionary Analysis: Facilitates study of evolutionary conservation and divergence of genetic networks across species.
• Pathway and Adaptation Identification: Aids identification of conserved pathways and species-specific adaptations through cross-species GI imputation.

Methodology:

Applies matrix factorization with composable models and incorporates kernels from side information such as protein-protein interaction networks; models were evaluated on matched genome-scale genetic interaction datasets from baker's yeast and fission yeast.