PfgPDI: Pocket feature-enabled graph neural network for protein-drug interaction prediction

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Protein-ligand interactions are crucial for drug discovery and understanding disease.
• Current methods lack accuracy due to incomplete feature representation and model adaptation issues.
• Accurate prediction of these interactions is vital for drug safety and efficacy.

Purpose of the Study:

• To develop a novel deep learning method for enhanced protein-ligand binding prediction.
• To improve the accuracy and reliability of predicting biomolecular interactions.
• To accelerate the drug discovery and development process.

Main Methods:

• A deep learning approach combining transformer networks and Graph Convolutional Networks (GCN).
• Transformer utilized for protein and SMILES sequence feature extraction, preventing local optima.
• Dilation convolutions and GCN applied to optimize pocket and SMILES features for classification.

Main Results:

• The proposed PfgPDI method demonstrated superior effectiveness compared to existing protein-ligand binding prediction techniques.
• Experimental results validate the model's capability in accurately predicting biomolecular interactions.
• The method shows significant potential for improving drug prediction accuracy.

Conclusions:

• The developed PfgPDI model offers a significant advancement in predicting protein-ligand interactions.
• This method is expected to accelerate new drug development and aid in drug testing.
• PfgPDI serves as a valuable tool for Research and Development engineers in the pharmaceutical industry.