GNN-SOM

GNN-SOM predicts the site-of-metabolism (SOM) within chemical compounds using Graph Neural Networks to characterize how metabolizing enzymes interact with substrates and affect metabolic pathways.

Key Features:

• Graph Neural Network Architecture: GNN-SOM uses a Graph Neural Network to classify atoms or bonds as potential sites-of-metabolism, capturing intramolecular relationships within molecular graphs.
• Comprehensive Training Data: The model is trained on enzymatic interaction data from the KEGG database encompassing all Enzyme Commission (EC) numbers.
• Performance Superiority: In comparative evaluations, GNN-SOM outperforms baseline machine-learning models in predicting SOMs for Cytochrome P450 (CYP) and non-CYP enzymes.

Scientific Applications:

• Extended Metabolic Models (EMMs): Predicted SOMs enable construction of extended metabolic models that incorporate enzyme promiscuity.
• Heterologous Synthesis Pathways: Prioritization of predicted enzymatic products supports design of heterologous synthesis pathways for synthetic biology and novel biosynthetic route construction.

Methodology:

Graph Neural Network classification of atoms or bonds trained on KEGG enzymatic interaction data spanning all EC numbers to learn patterns of metabolic activity across enzyme classes.