CoCoBind: Consistency-Contrastive Multitask Learning for RNA-Ligand Interaction and Binding Site Prediction.
TL;DR
CoBind is presented, a multitask deep learning framework that jointly predicts RNA–compound interactions and nucleotide-level binding-site probabilities within a unified architecture and provides complementary nucleotide-level binding-site localization, supporting a site-aware view of RNA–ligand recognition under distribution shift.
Abstract
RNA-targeted small-molecule therapeutics are of growing clinical interest, but computational modeling of RNA-ligand recognition remains challenging because of limited curated data, conformational heterogeneity, and the close coupling between interaction prediction and binding-site localization. We present CoCoBind, a multitask deep learning framework that jointly predicts RNA-compound interactions and nucleotide-level binding-site probabilities within a unified architecture. CoCoBind integrates pretrained RNA and molecular representation models with cross-modal cross-attention, a Noisy-OR-based consistency constraint, and contrastive alignment. On the DeepRNA-DTI benchmark, CoCoBind shows its clearest interaction-prediction gain in the most stringent Unseen Both setting and provides complementary nucleotide-level binding-site localization, supporting a site-aware view of RNA-ligand recognition under distribution shift. Structure-level analyses further support the localization of ligand-proximal RNA pocket neighborhoods. CoCoBind thus provides a site-aware computational framework for RNA-ligand interaction modeling and AI-enabled hit prioritization in RNA-targeted small-molecule discovery.