Protein ligand binding site prediction using graph transformer neural network

Area of Science:

• Computational chemistry and cheminformatics
• Structural biology and bioinformatics
• Artificial intelligence in drug discovery

Background:

• Ligand binding site prediction is vital for structure-based drug discovery.
• Existing geometry-based and machine learning methods have limitations in accuracy.
• Improving prediction performance is essential for efficient drug development.

Purpose of the Study:

• To develop and evaluate a novel approach for ligand binding site prediction.
• To leverage graph transformer neural networks for ranking prediction results.
• To investigate the impact of training dataset size on prediction performance.

Main Methods:

• Utilized a graph transformer neural network architecture.
• Integrated the neural network with a geometry-based pocket detection method.
• Created and employed a larger training dataset than conventionally used (e.g., sc-PDB).

Main Results:

• The graph transformer-based method effectively ranked results from the geometry-based method.
• A larger training dataset showed a positive correlation with improved prediction performance.
• The combined approach demonstrated enhanced accuracy in ligand binding site prediction.

Conclusions:

• Graph transformer neural networks offer a promising avenue for improving ligand binding site prediction.
• Increasing the size of the training dataset is crucial for enhancing model performance.
• This approach has the potential to accelerate structure-based drug discovery pipelines.