Complex machine learning model needs complex testing: Examining predictability of molecular binding affinity by a...

Area of Science:

• Computational chemistry
• Machine learning in drug discovery
• Bioinformatics

Background:

• Drug discovery relies on high-throughput screening, facing challenges with vast chemical spaces.
• Machine learning offers efficient computational methods for processing large compound libraries.

Purpose of the Study:

• To develop and assess a graph-based deep neural network for predicting protein-drug binding affinity.
• To evaluate model generalization using rigorous testing and cross-validation strategies.

Main Methods:

• Representing proteins and drugs as graphs processed by separate graph sub-networks.
• Training the neural network on PDBbind and RCSB Protein Data Bank datasets.
• Employing six distinct data partitioning strategies with k-fold cross-validation for performance assessment.

Main Results:

• The graph-based deep neural network shows predictive power for protein-drug binding affinity.
• Model performance was evaluated across various data split strategies, highlighting the need for comprehensive assessment.
• Code availability facilitates reproducibility and further research.

Conclusions:

• Graph-based deep learning provides an effective approach for predicting protein-drug binding affinity.
• Thorough validation strategies are essential for assessing model generalization in drug discovery.
• The developed model aids in navigating chemical spaces for identifying potential drug candidates.