GRIPHIN: grids of pharmacophore interaction fields for affinity prediction

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Pharmacophores are essential for characterizing protein-ligand interactions.
• Accurate binding affinity prediction is crucial for drug discovery.
• Current deep learning models often lack interpretability.

Purpose of the Study:

• To develop a hybrid deep learning framework for binding affinity prediction.
• To investigate the sufficiency of pharmacophoric representations for this task.
• To enhance model interpretability using attribution methods.

Main Methods:

• A hybrid framework combining pharmacophoric maps of protein binding sites and graph-based ligand representations was developed.
• A deep learning model was trained using these hybrid representations.
• Integrated gradients were applied for attribution analysis.

Main Results:

• The proposed hybrid method achieved performance comparable to state-of-the-art models.
• The model demonstrated interpretability by attributing predictions to specific pharmacophoric features.
• The study confirmed the potential of pharmacophoric representations in deep learning for affinity prediction.

Conclusions:

• Pharmacophore-based deep learning models can effectively predict binding affinity.
• The hybrid approach offers a balance of predictive accuracy and interpretability.
• This framework advances the application of pharmacophores in computational drug discovery.