deepGraphh: AI-driven web service for graph-based quantitative structure-activity relationship analysis

Area of Science:

• Chemoinformatics
• Computational Chemistry
• Drug Discovery

Background:

• Artificial intelligence (AI) accelerates chemical space exploration.
• Compound representation is critical for quantitative structure-activity relationship (QSAR) analysis.
• Graph-based methods offer advantages over traditional descriptors but require specialized expertise.

Purpose of the Study:

• Introduce deepGraphh, an end-to-end web service for graph-based QSAR model generation.
• Provide a user-friendly interface for classification and regression tasks.
• Facilitate model tuning, generation, cross-validation, and testing of query molecules.

Main Methods:

• Implemented four established graph-based methods: graph convolution network, graph attention network, directed acyclic graph, and Attentive FP.
• Developed a graphical user interface (GUI) for accessible model parameter configuration.
• Integrated capabilities for cross-validation and testing of user-supplied molecules.

Main Results:

• DeepGraphh models demonstrate performance comparable to descriptor-based machine learning techniques.
• Successfully predicted blood-brain barrier permeability for human and microbiome-generated metabolites.
• Validated the utility of graph-based methods in chemoinformatics applications.

Conclusions:

• DeepGraphh provides a comprehensive, one-stop web service for graph-based QSAR analysis.
• Lowers the barrier to entry for utilizing advanced graph-based modeling techniques.
• Empowers researchers in drug discovery and chemical biology through accessible AI tools.