Deep Learning for Drug Development: Using CNNs in MIA-QSAR to Predict Plasma Protein Binding of Drugs

Area of Science:

• Pharmacokinetics
• Drug Development
• Computational Chemistry

Background:

• Plasma protein binding (PPB) is a critical factor influencing drug efficacy and safety.
• Accurate prediction of PPB is essential for successful drug development.
• Traditional methods for PPB prediction can be time-consuming and resource-intensive.

Purpose of the Study:

• To develop and validate a deep learning model for predicting plasma protein binding (PPB).
• To leverage convolutional neural networks (CNNs) for extracting molecular features relevant to PPB.
• To assess the accuracy and generalizability of the developed model in drug development.

Main Methods:

• A convolutional neural network (CNN) was utilized to extract 10 numerical features from the molecular structures of 100 drugs.
• These extracted molecular features served as input for a feedforward network to predict PPB.
• The model's performance was evaluated using training, external validation, and cross-validation metrics.

Main Results:

• The CNN successfully extracted key molecular features influencing PPB.
• High prediction accuracy was achieved, with an R² train of 0.89 and an external validation R² of 0.931.
• A low cross-validation mean squared error (CV-MSE) of 0.0213 was obtained, indicating model robustness.

Conclusions:

• Deep learning techniques, particularly CNNs, are effective for predicting plasma protein binding (PPB).
• The developed model demonstrates significant potential for enhancing pharmacokinetic predictions in drug development.
• This AI-driven approach offers a promising tool for improving drug efficacy and safety assessments.