Prediction of drug protein interactions based on variable scale characteristic pyramid convolution network

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Accurate drug-target binding affinity prediction is crucial for efficient drug screening.
• Deep learning methods, particularly multilayer convolutional neural networks (CNNs), are widely used for this task.
• A limitation of deep CNNs is the potential loss of semantic information from low-level features with increasing network depth, impacting prediction accuracy.

Purpose of the Study:

• To propose a novel deep learning method, Pyramid Network Convolution Drug-Target Binding Affinity (PCNN-DTA), to enhance drug-target binding affinity prediction.
• To address the issue of low-level feature information loss in deep CNNs for affinity prediction.

Main Methods:

• The proposed PCNN-DTA method utilizes a feature pyramid network (FPN) architecture.
• It fuses features extracted from multiple layers of a multilayer CNN.
• Input data includes simplified molecular input system (SMILES) strings for compounds and amino acid sequences for proteins.

Main Results:

• The PCNN-DTA method demonstrated superior performance compared to existing regression prediction methods based on CNNs.
• Evaluations were conducted on three benchmark datasets: KIBA, Davis, and Binding DB.
• The fusion of multi-level features effectively preserved low-level information, leading to improved prediction accuracy.

Conclusions:

• The PCNN-DTA method offers a significant advancement in predicting drug-target binding affinity.
• Its feature fusion strategy effectively mitigates information loss in deep learning models.
• PCNN-DTA shows strong potential for application in drug discovery and development.