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

RNA-binding proteins (RBPs) often recognize complex RNA secondary structure motifs, and these interactions can vary between cell lines. However, no current computational model can explain the dynamic behavior of RNA sequence-structure motifs across multiple cell lines and link these dynamics to the functional impact of genetic variation. Here we show BRIDGE, an end-to-end unified model that bridges sequence-structure motifs with the functional effects of noncoding genetic variants, thereby enabling genome-wide analysis of dynamic RNA-protein interaction profiles. Our evaluations demonstrate that BRIDGE outperforms existing state-of-the-art methods in static single cell line predictions. Without any retraining, the framework transfers to unseen cellular contexts, thereby revealing a conserved yet adaptable grammar of RNA recognition. Attention-guided interpretation identified 3,571 integrative motifs, including motifs associated with splicing regulation. Moreover, genome-wide in-silico perturbation provides an interpretable view of RBP-binding disruption across variant classes, including splice-region and pathogenic alleles. RNA-binding proteins recognize RNA structures that vary across cell types and genetic variation. Here, the authors present BRIDGE, a model predicting RNA-protein interactions across contexts and explaining variant effects genome-wide.