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

Abstract

Motivation While traditionally utilized for identifying site-specific metabolic activity within a compound to alter its interaction with a metabolizing enzyme, predicting the Site-of-Metabolism (SOM) is essential in analyzing the promiscuity of enzymes on substrates. The successful prediction of SOMs and the relevant promiscuous products has a wide range of applications that include creating extended metabolic models that account for enzyme promiscuity and the construction of novel heterologous synthesis pathways. There is therefore a need to develop generalized methods that can predict molecular SOMs for a wide range of metabolizing enzymes. Results This paper develops a Graph Neural Network (GNN) model for the classification of an atom (or a bond) being an SOM. Our model, GNN-SOM, is trained on enzymatic interactions, available in the KEGG database, that span all enzyme commission numbers. We demonstrate that GNN-SOM consistently outperforms baseline Machine Learning (ML) models, when trained on all enzymes, on Cytochrome P450 (CYP) enzymes, or on non-CYP enzymes. We showcase the utility of GNN-SOM in prioritizing predicted enzymatic products due to enzyme promiscuity for two biological applications: the construction of Extended Metabolic Models (EMMs) and the construction of synthesis pathways. Availability A python implementation of the trained SOM predictor model can be found at https://github.com/HassounLab/GNN-SOM Supplementary information Not applicable

Molecular structure discovery for untargeted metabolomics using biotransformation rules and global molecular networking

“…Therefore, a systematic approach is needed to prioritize the suggested chemical identities of the suspect. For this, we use GNN-SOM [10], a tool that employs a graph neural network to predict the likelihood of each atom in a molecule being the target of a biotransformation operation. GNN-SOM was trained on enzymatic interactions from the KEGG database to classify each atom within a graph representation of a molecule as the site-of-metabolism for a given enzyme.…”

“…To do so, BAM makes use of PROXI-MAL2 [9] to apply the biotransformation rules to the anchor molecule, identifying suitable reaction centers for the proposed structural changes. Next, the graph neural network-based GNN-SOM tool is used to predict the site-of-metabolism and rank these putative derivatives based on the likelihood of each atom being a reaction center [10].…”

Enzyme promiscuity approaches to predict metabolic pathways in the human gut microbiota.

“…Bigger fragments correspond to more specific definitions of the rules and, hence, promiscuity is more limited. The tuning of specificity and parameters is a challenging and important process of the rule-based approaches and, when the parameters are not well tuned, they can lead to infeasible biotransformations (Ni et al, 2021;Porokhin et al, 2023).…”

“…Despite these early advances, the definition of features to extract and apply the promiscuous activity of the enzymes is still challenging, leaving the door open for further improvements. In addition, it has been investigated the possibility to combine rules-based methods and machine learning approaches, so as to enhance predictions and enrich results (Porokhin et al, 2023;Zhang and Sousa, 2005).…”

“…They can be classified following several criteria (Wang et al, 2017). For example, Nakamura Finally, a more general classification of enzyme promiscuity divides methods into those based on reaction rules and machine learning (Porokhin et al, 2023;Zhang and Sousa, 2005). Although in this doctoral thesis we will concentrate on rule-based methods, we describe these two approaches, highlighting their beneficial aspects and limitations.…”