Using graph neural networks for site-of-metabolism prediction and its applications to ranking promiscuous enzymatic products

• DOI: 10.1093/bioinformatics/btad089
• Publication Date: 2023-02-15T05:42:03Z
• Authors: Vladimir Porokhin; Li-Ping Liu; Soha Hassoun
• Keywords: Python; KEGG

Abstract Motivation While traditionally utilized for identifying site-specific metabolic activity within a compound to alter its interaction with metabolizing enzyme, predicting the site-of-metabolism (SOM) is essential in analyzing promiscuity of enzymes on substrates. The successful prediction SOMs and relevant promiscuous products has wide range applications that include creating extended models (EMMs) account enzyme construction novel heterologous synthesis pathways. There therefore need develop generalized methods can predict molecular enzymes. Results This article develops Graph Neural Network (GNN) model classification an atom (or bond) being SOM. Our model, GNN-SOM, trained enzymatic interactions, available KEGG database, span all commission numbers. We demonstrate GNN-SOM consistently outperforms baseline machine learning models, when enzymes, Cytochrome P450 (CYP) or non-CYP showcase utility prioritizing predicted due two biological applications: EMMs Availability implementation A python SOM predictor be found at https://github.com/HassounLab/GNN-SOM. Supplementary information data are Bioinformatics online.