Graph convolutional neural networks improved target-specific scoring functions for cGAS and kRAS in virtual screening

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning in bioinformatics

Background:

• Traditional virtual screening methods often rely on limited empirical scoring functions.
• Machine learning-based target-specific scoring functions show promise for improving virtual screening accuracy.
• Extrapolation performance is key for the broad applicability of these scoring functions.

Purpose of the Study:

• To improve the extrapolation ability of target-specific scoring functions using molecular graph and convolutional neural networks.
• To enhance the accuracy and robustness of virtual screening for drug discovery.
• To evaluate the performance of graph convolutional networks for predicting molecular activity.

Main Methods:

• Developed target-specific scoring functions using machine learning, including graph convolutional networks.
• Applied rigorous data screening and feature extraction for cGAS and kRAS proteins.
• Constructed and compared traditional machine learning and deep learning models.

Main Results:

• Target-specific scoring functions significantly outperformed generic ones.
• Graph convolutional networks demonstrated remarkable robustness and accuracy in predicting molecular activity.
• The models showed superiority in screening efficiency and accuracy for cGAS and kRAS targets.

Conclusions:

• Graph convolutional networks can be generalized for predicting heterogeneous data based on learned molecular-protein binding patterns.
• Target-specific scoring functions, particularly those using graph convolutional networks, hold significant potential for structure-based virtual screening.
• This approach enhances screening efficiency and accuracy in drug discovery pipelines.