deepGraphh

deepGraphh applies graph-based neural network methods for chemoinformatics to perform QSAR modeling and predict molecular properties, including blood-brain barrier permeability for human and microbiome-generated metabolites.

Key Features:

• End-to-end modeling: Provides model generation, parameter tuning, cross-validation, and testing from input molecules to model evaluation.
• Graph-based methods: Implements Graph Convolution Network (GCN), Graph Attention Network (GAT), Directed Acyclic Graph, and Attentive FP.
• Query molecule testing: Supports testing and evaluation using user-supplied query molecules.
• Comparative performance: Delivers performance comparable to descriptors-based machine learning techniques for QSAR tasks.

Scientific Applications:

• QSAR modeling: Applied to quantitative structure-activity relationship analysis and prediction of molecular properties.
• Blood-brain barrier permeability prediction: Used to predict blood-brain barrier permeability for human and microbiome-generated metabolites relevant to drug discovery and development.

Methodology:

Transforms chemical compounds into computational-compatible features using artificial intelligence and applies graph-based neural networks (GCN, GAT, Directed Acyclic Graph, Attentive FP), with parameter tuning, cross-validation, and testing; uses graph-based representations instead of traditional fingerprints or descriptors.