Evaluation of Small-Molecule Binding Site Prediction Methods on Membrane-Embedded Protein Interfaces

Area of Science:

• Computational biology
• Drug discovery
• Structural bioinformatics

Background:

• Drug molecules increasingly target the protein-membrane interface in membrane proteins.
• Accurate ligand binding site identification is essential but challenging for membrane proteins.
• Existing computational methods need evaluation for predicting binding sites within the membrane-embedded regions.

Purpose of the Study:

• To assess the performance of various computational methods for predicting ligand binding sites in the intramembrane region of membrane proteins.
• To compare these methods against their performance on soluble proteins.
• To identify the most effective methods for drug discovery targeting membrane proteins.

Main Methods:

• Compiled datasets of G-protein coupled receptor (GPCR) and ion channel-ligand complexes.
• Evaluated geometry-based, energy probe-based, machine learning, and deep learning methods.
• Assessed performance using center-to-center distance (DCC) and discretized volume overlap (DVO).

Main Results:

• DeepPocket, PUResNetV2.0, and ConCavity ranked highest for GPCRs.
• DeepPocket, PUResNetV2.0, and FTSite showed the best performance for ion channels.
• All tested methods demonstrated lower performance on membrane proteins compared to soluble proteins.

Conclusions:

• This study offers an overview of current computational binding site prediction method performance for membrane protein interfaces.
• Deep learning methods show promise but further development is needed for accurate prediction in protein-membrane regions.
• Improved computational tools are essential for advancing drug discovery targeting membrane proteins.