graphDelta: MPNN Scoring Function for the Affinity Prediction of Protein-Ligand Complexes.

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning in structural biology

Background:

• Accurate prediction of protein-ligand binding affinity is crucial for drug discovery.
• Existing methods often struggle with the complexity of molecular interactions.

Purpose of the Study:

• To develop a novel graph-convolutional neural network (GCN) model for predicting binding constants.
• To evaluate the model's performance using multiple binding affinity metrics.

Main Methods:

• Utilized multi-task learning to predict dissociation constant (Kd), inhibition constant (Ki), and half maximal inhibitory concentration (IC50).
• Trained the GCN model rigorously on the comprehensive PDBbind dataset.
• Employed graph convolutional networks to capture complex structural features of protein-ligand complexes.

Main Results:

• Achieved a Pearson correlation coefficient of 0.87 and a Root Mean Square Error (RMSE) of 1.05 pK units.
• The GCN model demonstrated superior performance compared to a 3D convolutional neural network (Kdeep).

Conclusions:

• Graph-convolutional neural networks offer a powerful approach for accurate binding constant prediction.
• The developed model shows significant potential for accelerating drug discovery and development pipelines.