Disobind: A sequence-based, partner-dependent contact map and interface residue predictor for intrinsically...

Area of Science:

• Computational Biology
• Protein Science
• Bioinformatics

Background:

• Intrinsically disordered proteins (IDPs) exhibit dynamic structures and diverse binding modes.
• Characterizing the interfaces of IDPs remains a significant experimental and computational challenge.
• Current prediction tools like AlphaFold struggle with IDP binding site accuracy.

Purpose of the Study:

• To develop a novel deep-learning method, Disobind, for predicting inter-protein contact maps and interface residues for IDPs.
• To improve the accuracy and efficiency of IDP interface prediction compared to existing methods.

Main Methods:

• Disobind utilizes sequence embeddings from the ProtT5 protein language model.
• The method predicts protein-protein contact maps and interface residues directly from amino acid sequences.
• Performance was evaluated against state-of-the-art interface predictors and AlphaFold models.

Main Results:

• Disobind significantly outperforms existing interface predictors for IDPs.
• Disobind demonstrates superior performance compared to AlphaFold multimer and AlphaFold3 across various confidence thresholds.
• Combining Disobind with AlphaFold-multimer predictions further enhances prediction accuracy.

Conclusions:

• Disobind offers a robust and accurate approach for characterizing IDP-mediated interactions without requiring structural or multiple sequence alignment data.
• The method's ability to consider binding partner context and rely solely on sequences makes it a valuable tool for IDP research.
• Disobind predictions can aid in localizing IDPs within large molecular assemblies and elucidating their functional roles.