A novel NLP-based method and algorithm to discover RNA-binding protein (RBP) motifs, contexts, binding preferences,...

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• RNA-binding proteins (RBPs) regulate mRNA processing, but their binding specificities and interactions are poorly understood.
• Existing computational methods for predicting RBP binding lack interpretability and fail to adequately address motif context and RBP-RBP interactions.
• There is a need for interpretable models to understand the contextual factors influencing RBP binding in vivo.

Purpose of the Study:

• To develop a novel, interpretable computational pipeline for predicting RBP binding specificity.
• To characterize RBP binding motifs and contexts, and to identify novel RBP-RBP interactions and their regulatory roles.

Main Methods:

• A Natural Language Processing (NLP)-based method was used to deconstruct RNA sequences into k-mers and flanking regions.
• RBP binding prediction was formulated as a weakly supervised Multiple Instance Learning problem.
• A deterministic motif discovery algorithm was developed for prediction interpretability, and feature integration was used to infer RBP-RBP interactions.

Main Results:

• The pipeline successfully recapitulated known RBP motifs, validating its predictive capability.
• Binding motifs and contexts were characterized for 71 RBPs in HepG2 cells and 74 RBPs in K562 cells, with many novel findings.
• Novel cooperative and competitive RBP-RBP interaction partners were proposed, along with hypotheses on their regulatory functions.

Conclusions:

• The developed framework provides a comprehensive approach to investigate RBP binding specificity determinants.
• The findings enhance understanding of RBP binding patterns, interactions, and regulatory functions.
• This work offers a valuable tool for future research in RNA biology and gene regulation.