Bridging sequence-structure motifs and genetic variants for genome-wide dynamic RNA-protein interaction profiling.

RNA-binding proteins interact with RNA secondary structure motifs, and these interactions vary across cell lines. Prior research has shown that RNA structure motifs influence RBP binding. However, no existing model captures how these motifs change dynamically. Genetic variants in noncoding regions can affect RNA structure and RBP binding. Current tools lack the ability to link motif dynamics to variant effects. This gap motivated the development of a unified framework. No prior work had resolved the genome-wide functional impact of such interactions. That uncertainty drove the need for a model that integrates structure and variant data. This paper introduces a novel approach to address these limitations.

The study aimed to develop a computational model that connects RNA sequence-structure motifs to the effects of noncoding genetic variants. The researchers sought to enable genome-wide profiling of RNA-protein interactions. They wanted to account for dynamic motif behavior across cell lines. Their goal was to create a framework that transfers across cellular contexts. The model needed to interpret RBP binding disruption. They also aimed to identify motifs linked to splicing regulation. The study focused on improving predictions compared to existing tools. It aimed to provide interpretable insights into variant effects on RNA structure.

The team developed BRIDGE, an end-to-end model that links RNA motifs to variant effects. They used sequence-structure motifs as input features. The model incorporates RNA secondary structure predictions. They trained it on RNA-protein interaction data from multiple cell lines. BRIDGE uses attention mechanisms to interpret motif importance. The framework does not require retraining for new cell lines. They evaluated performance using static single-cell line predictions. The model was tested on its ability to transfer to unseen contexts.

BRIDGE outperformed existing methods in static predictions. The model achieved high accuracy without retraining. It transferred well to new cellular contexts. Attention analysis revealed 3,571 integrative motifs. Some motifs were linked to splicing regulation. In-silico perturbation showed variant effects on RBP binding. The model identified disruptions from splice-region alleles. It also captured effects from pathogenic variants. These findings suggest BRIDGE captures conserved motif patterns.

The authors proposed that BRIDGE provides a unified framework for RNA motif and variant analysis. They suggested the model enables dynamic interaction profiling. The framework transfers across cell lines without retraining. The study demonstrated BRIDGE's ability to interpret variant effects. Attention analysis revealed conserved motif patterns. The model's performance exceeded existing tools. The findings suggest RNA recognition follows an adaptable grammar. The authors emphasized BRIDGE's utility for genome-wide analyses.