A Deep Learning-Based Quantitative Structure-Activity Relationship System Construct Prediction Model of Agonist and...

Area of Science:

• Computational Chemistry
• Toxicology
• Artificial Intelligence

Background:

• Quantitative structure-activity relationship (QSAR) methods, particularly those employing deep learning (DL), have advanced molecular design and safety assessment.
• Previous work introduced DeepSnap-DL, a DL-based QSAR system that uses images of 3D chemical structures to predict molecular initiation events (MIEs) linked to adverse toxicological outcomes.
• A limitation of the original DeepSnap-DL system was its time-consuming nature.

Purpose of the Study:

• To enhance the DeepSnap-DL system for improved efficiency and prediction performance in QSAR analysis.
• To optimize the parameters involved in image generation from 3D chemical structures and the subsequent deep learning model.
• To establish a more powerful tool for computer-aided molecular design and toxicological assessment.

Main Methods:

• Developed an improved DeepSnap-DL system integrating 3D chemical structure to image generation, deep learning, and statistical performance calculation.
• Optimized key parameters including image depiction angles, data-splitting strategies, and deep learning hyperparameters.
• Evaluated prediction performance for three models predicting agonists or antagonists of MIEs.

Main Results:

• The improved DeepSnap-DL system demonstrated high prediction performance for MIE-related models.
• Optimization of parameters led to significant improvements in the accuracy and efficiency of the QSAR predictions.
• The refined system effectively predicts the activity of molecules concerning MIEs.

Conclusions:

• The optimized DeepSnap-DL system offers a powerful and efficient approach for QSAR analysis.
• This enhanced system is a valuable tool for computer-aided molecular design and chemical safety evaluations.
• The study highlights the potential of advanced AI techniques in accelerating drug discovery and toxicology prediction.