Protein-ligand binding region prediction (PLB-SAVE) based on geometric features and CUDA acceleration

Area of Science:

• Computational biology
• Bioinformatics
• Structural biology

Background:

• Protein-ligand interactions are crucial for cellular functions and molecular recognition.
• Predicting protein binding regions aids in understanding these interactions and facilitates drug discovery.
• In silico geometrical analysis of protein surfaces is a key step in bioinformatics for drug design.

Purpose of the Study:

• To develop novel software, PLB-SAVE, for predicting protein binding regions using geometrical features.
• To enhance the efficiency and accuracy of binding region prediction through parallel processing technology.

Main Methods:

• PLB-SAVE extracts geometrical constructs (solid angles) from protein surface atoms.
• It identifies and ranks representative clusters and anchors based on solid angles.
• Cavity depth and volume are calculated and combined with weighting coefficients to rank potential binding regions.

Main Results:

• PLB-SAVE demonstrated superior accuracy (94.3% for unbound, 95.5% for bound proteins) compared to SiteHound and MetaPocket2.0.
• A tenfold cross-validation process validated the prediction accuracy.
• The parallel processing architecture resulted in an average 160-fold increase in computational speed.

Conclusions:

• PLB-SAVE offers a robust in silico method for protein-ligand binding region prediction, crucial for structure-based drug design.
• The software's geometrical approach improves the efficacy of biological experiments in drug development.
• The method's applicability extends to predicting carbohydrate-antibody interactions for vaccine development.