A deep learning method for predicting interactions for intrinsically disordered regions of proteins

Intrinsically disordered proteins or regions (IDPs/IDRs) exist as ensembles of conformations in the monomeric state. Upon binding to a partner, they adopt various binding modes, ranging from becoming ordered upon binding, to binding in a multivalent manner, to remaining fuzzy in the bound state. Moreover, they can adopt different binding modes depending on the partner. Thus, characterizing the interfaces of IDRs in complexes is challenging experimentally and computationally. Alphafold-multimer and Alphafold3, the state-of-the-art structure prediction methods, are less accurate at predicting binding sites of IDRs in complexes, at their benchmarked confidence cutoffs. However, their performance improves upon lowering the confidence cutoffs. Here, we developed Disobind, a deep-learning method that predicts inter-protein contact maps and interface residues for an IDR and a partner protein, given their sequences. It performs better than AlphaFold-multimer and AlphaFold3 across multiple confidence cutoffs. Combining the Disobind and AlphaFold-multimer predictions further improves the performance. In contrast to most current methods, Disobind considers the context of the binding partner, does not require the structure of either protein, and is not dependent on multiple sequence alignments. Its predictions can be used to localize IDRs in integrative structures of large assemblies, characterize protein-protein interactions involving IDRs, and modulate IDR-mediated interactions.