IRIS Integrates Sparse Sequence, Experimental, and AI-Predicted Structures for Protein-RNA Affinity Prediction and Motif Discovery
Cisneros de la Rosa, E.; Zhang, Y.; Lin, X.
Show abstract
Protein-RNA interactions are fundamental to numerous cellular processes, yet quantitatively characterizing their binding specificity remains a major challenge. We present IRIS (Integrative RNA-protein interaction prediction Informed by Structure and sequence), a biophysical framework that combines residue-level sequence and structural features to predict binding affinities and identify binding motifs. Applied across different protein-RNA systems, IRIS accurately predicts the effects of multiple RNA mutations, with performance further improved by incorporating additional high-affinity binders into training. By integrating predicted structural complexes, IRIS reveals alternative binding modes not observed in experimental structures, extends applicability to systems lacking experimental protein-RNA complexes, and generates a library of favorable RNA-binding motifs at protein-RNA interfaces. Collectively, these results establish IRIS as a versatile framework that leverages increasingly accurate structural predictions to enable quantitative modeling and rational engineering of protein-RNA interactions.
Matching journals
The top 6 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Generative and predictive neural networks for the design of functional RNA molecules 96%
- Secondary Structure Prediction for RNA Sequences Including N6-methyladenosine 96%
- The solution structure of Dead End bound to AU-rich RNA reveals an unprecedented mode of tandem RRM-RNA recognition required for mRNA regulation 96%
Similar papers in this journal
Similar papers in this journal
Similar papers in this journal
- The importance of input sequence set to consensus-derived proteins and their relationship to reconstructed ancestral proteins 94%
- COLLAPSE: A representation learning framework for identification and characterization of protein structural sites 94%
- Neural Network-Derived Potts Models for Structure-Based Protein Design using Backbone Atomic Coordinates and Tertiary Motifs 93%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.