Protein-small molecule binding site prediction based on a pre-trained protein language model with contrastive...

Area of Science:

• Computational biology
• Structural bioinformatics
• Drug discovery

Background:

• Predicting protein-small molecule binding sites is crucial for drug design but challenging for proteins lacking experimental structures.
• Existing methods struggle with proteins without published crystal structures, including intrinsically disordered proteins (IDPs).

Purpose of the Study:

• To develop and validate CLAPE-SMB, a computational tool for accurate prediction of protein-small molecule binding sites.
• To enable structure-guided drug design for proteins lacking experimental structural data.

Main Methods:

• CLAPE-SMB integrates a pre-trained protein language model with contrastive learning.
• The model was trained and tested on diverse datasets, including SJC, UniProtSMB, and an intrinsically disordered protein (IDP) dataset.
• Performance was evaluated using the Matthews Correlation Coefficient (MCC).

Main Results:

• CLAPE-SMB achieved an MCC of 0.529 on the SJC dataset.
• It reached an MCC of 0.699 on the UniProtSMB dataset and 0.815 on the IDP dataset.
• Case studies demonstrated CLAPE-SMB's potential to aid drug design for specific proteins.

Conclusions:

• CLAPE-SMB accurately predicts small molecule binding sites, particularly for proteins without experimental structures like IDPs.
• The model's adaptability across datasets makes it a valuable tool for drug design and understanding protein-small molecule interactions.
• The freely available code and datasets facilitate further research and application.