iONMF

iONMF models and predicts protein-RNA interactions by integrating multiple sources of experimental data with genome annotations, gene functions, RNA sequences, and RNA structures using integrative orthogonality-regularized nonnegative matrix factorization (iONMF) to discover non-overlapping, class-specific RNA-binding patterns that vary in strength.

Key Features:

• Integrative orthogonality-regularized NMF: Applies nonnegative matrix factorization with an orthogonality constraint (iONMF) to extract class-specific, non-overlapping binding patterns.
• Orthogonality regularization: Incorporates an orthogonality constraint that reduces the size of the factor model by half and enhances discrimination of patterns across data sources.
• Multi-source data integration: Integrates experimental RBP data with genome annotations, gene functions, RNA sequences, and RNA structures.
• CLIP compendium: Uses a compendium including 31 CLIP experiments on 19 RBPs involved in splicing and 3'UTR processing.
• Predictive factors: Identifies influences on binding such as position of RNA structure and sequence motifs, co-binding of RBPs, and gene region types.
• Protein-specific pattern discovery: Reveals protein-specific patterns consistent with experimentally determined properties of RBPs such as hnRNPs, U2AF2, ELAVL1, TDP-43, and FUS.

Scientific Applications:

• Post-transcriptional regulation analysis: Improves modeling of mechanisms including RNA stability, splicing, transport, and polyadenylation by predicting RBP interaction sites.
• RBP binding site prediction: Enhances accuracy in identifying RNA sites bound by RBPs across gene regions.
• Regulatory network exploration: Facilitates study of co-binding relationships and class-specific binding patterns to investigate regulatory networks and dysregulation in disease.

Methodology:

Applies integrative orthogonality-regularized nonnegative matrix factorization with an orthogonality constraint to integrated datasets (CLIP experiments, genome annotations, gene functions, RNA sequences, RNA structures) to extract discriminative, class-specific binding factors such as sequence motifs, structural positions, co-binding signals, and gene-region effects.