CENsible: Interpretable Insights into Small-Molecule Binding with Context Explanation Networks.

Area of Science:

• Computational chemistry
• Structural biology
• Machine learning in drug discovery

Background:

• Accurate prediction of small-molecule binding affinity is crucial for drug discovery.
• Existing machine learning models often lack interpretability, hindering optimization efforts.

Purpose of the Study:

• To present CENsible, a novel and interpretable deep learning approach for assessing protein/ligand binding affinity.
• To demonstrate the ability of CENsible to distinguish active from inactive compounds.

Main Methods:

• Utilizing context explanation networks (CENs), a deep convolutional neural network architecture.
• Predicting contributions of precalculated terms to the overall binding affinity for protein/ligand complexes.

Main Results:

• CENsible effectively distinguishes active from inactive compounds across various systems.
• The model provides interpretability by identifying the contribution of each term to the final prediction.

Conclusions:

• CENsible offers an interpretable alternative to existing machine learning scoring functions.
• The interpretability of CENsible has direct implications for guiding lead optimization in drug development.