Natural language-based representation and modeling of RBP binding

Area of Science:

• Computational Biology
• Molecular Biology
• Genomics

Background:

• RNA-binding proteins (RBPs) are crucial for regulating gene expression via RNA.
• Current methods for characterizing RBP binding patterns are limited by interpretability and context-insensitivity.
• The precise role of sequence context in RBP binding specificity is not well understood.

Purpose of the Study:

• To develop a novel computational framework for deciphering RBP binding patterns.
• To investigate the contribution of RNA sequence context to RBP binding specificity.
• To create an interpretable method for identifying key RNA regions involved in RBP binding.

Main Methods:

• A natural language-based representation was developed for RNA sequences (lexical, syntactic, semantic).
• RNA sequences were decomposed into regions containing k-mers and flanking contexts.
• A Multiple Instance Learning (MIL) framework with 'iterative relabeling' was used to predict RBP binding.

Main Results:

• The approach successfully models RNA sequences using linguistic concepts.
• Key RNA regions contributing to RBP binding were identified.
• The method demonstrated accuracy and interpretability in predicting RBP binding.

Conclusions:

• The novel approach provides a comprehensive and interpretable method for RBP binding pattern analysis.
• Understanding RBP binding context is essential for transcriptome regulation.
• This framework advances computational methods for studying RNA-protein interactions.