Orthogonal matrix factorization enables integrative analysis of multiple RNA binding proteins

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• RNA binding proteins (RBPs) regulate gene expression post-transcriptionally.
• Accurate modeling of protein-RNA interactions requires integrating diverse data, including genomic annotation, gene function, and RNA sequence/structure.
• Existing matrix factorization methods often identify only strong patterns, missing crucial weaker interactions.

Purpose of the Study:

• To develop a novel method for integrating multiple data sources to model protein-RNA interactions.
• To discover non-overlapping RNA binding patterns with varying strengths.
• To improve the prediction of RNA binding sites by capturing complex interaction landscapes.

Main Methods:

• Developed integrative orthogonality-regularized nonnegative matrix factorization (iONMF).
• Integrated a large compendium of 31 CLIP experiments across 19 RBPs involved in splicing and 3'UTR processing.
• Utilized orthogonality constraints to enhance model efficiency and predictive performance compared to standard NMF.

Main Results:

• iONMF successfully integrated multiple data sources, improving the prediction accuracy of RNA binding sites.
• Key predictive factors identified include RNA structure and sequence motifs, RBP co-binding, and gene region.
• The method revealed protein-specific binding patterns consistent with known RBP properties.

Conclusions:

• iONMF provides a powerful framework for modeling complex protein-RNA interactions.
• The integration of diverse data and the discovery of varying strength patterns enhance predictive capabilities.
• The developed method and data compendium offer valuable resources for understanding RNA binding protein functions.