GLORYx: Prediction of the Metabolites Resulting from Phase 1 and Phase 2 Biotransformations of Xenobiotics

Abstract

Predicting the structures of metabolites formed in humans can provide advantageous insights for the development of drugs and other compounds. Here we present GLORYx, which integrates machine learning-based site of metabolism (SoM) prediction with reaction rule sets to predict and rank the structures of metabolites that could potentially be formed by phase 1 and/or phase 2 metabolism. GLORYx extends the approach from our previously developed tool GLORY, which predicted metabolite structures for cytochrome P450-mediated metabolism only. A robust approach to ranking the predicted metabolites is attained by using the SoM probabilities predicted by the FAME 3 machine learning models to score the predicted metabolites. On a manually curated test data set containing both phase 1 and phase 2 metabolites, GLORYx achieves a recall of 77% and an area under the receiver operating characteristic curve (AUC) of 0.79. Separate analysis of performance on a large amount of freely available phase 1 and phase 2 metabolite data indicates that achieving a meaningful ranking of predicted metabolites is more difficult for phase 2 than for phase 1 metabolites. GLORYx is freely available as a web server at https://nerdd.zbh.uni-hamburg.de/ and is also provided as a software package upon request. The data sets as well as all the reaction rules from this work are also made freely available.

Figure 1

Comparison of the metabolite data from MetXBioDB and DrugBank, in terms of parent molecules. (A) Distribution of molecular weight. (B) Distribution of clog P. (C) Histogram of the number of metabolites per parent molecule in terms of percentage of parent molecules. (D) Comparison of the chemical space of the parent molecules from MetXBioDB and DrugBank using PCA calculated using 44 physicochemical descriptors. The percentage of the total variance explained by each of the first two principal components is included in the axis labels.

Figure 2

Rank-based ROC curves for the evaluation of metabolite prediction performance on the reference data set. The ranks are calculated based on the priority scores of the predicted metabolites for each parent molecule. (A) Comparison of GLORYx, which scores its predicted metabolites based on predicted SoM probabilities, to SyGMa, which uses reaction probability scores, for phase 1 metabolite prediction. Weighted rules refer to the weighting of the SoM probability-based score based on whether the reaction type is designated common or uncommon. (B) Comparison of the ranking performance of GLORYx with different scoring approaches and rule sets as well as direct comparison to SyGMa’s performance, for phase 2 metabolite prediction. The scoring approach that is based on both SoM probability and reaction probability is achieved by a simple multiplication of the two components. (C) Comparison of the ranking performance of GLORYx for combined prediction of metabolites for phases 1 and 2 metabolism, using different SoM prediction approaches to score the predicted metabolites. In both cases, the score is based on predicted SoM probability with weighting according to reaction type, and the rule set is made up of the final phase 1 rule set (SyGMa and GLORY rules) and final phase 2 rule set (SyGMa and GSH conjugation rules).

Figure 3

Workflow of phase 2 metabolite prediction using reaction type-specific SoM models to score and rank the predicted metabolites. The reaction type-specific SoM models (“UGT”, “GST”, “SULT”, “NAT”, “MT”) are used instead of the general phase 2 SoM model (P2) to score the products of the relevant reactions for all molecules in which all of the reaction type-specific models are able to make a prediction. The green arrows indicate the molecules that were predicted successfully by the relevant reaction type-specific SoM model. If one or more of the reaction type-specific models cannot make predictions for a given molecule, then that molecule additionally follows the path of the black arrows, followed by a deduplication of predictions. The “UGT” model covers glucuronidation and glycosylation reactions, the “GST” model covers GSH and RSH conjugations, the “SULT” model covers sulfonations, the “NAT” model covers acetylations and acylations, and the “MT” model covers methylation reactions. The “other phase 2 rules” refer to the rules that are neither glucuronidation, GSH conjugation, sulfonation, acetylation, or methylation rules.

Figure 4

ROC curves for GLORYx and SyGMa representing ranking performance on the test set based on the (A) ranks and (B) scores of the predicted metabolites.