GLORY: Generator of the Structures of Likely Cytochrome P450 Metabolites Based on Predicted Sites of Metabolism

Abstract

Computational prediction of xenobiotic metabolism can provide valuable information to guide the development of drugs, cosmetics, agrochemicals, and other chemical entities. We have previously developed FAME 2, an effective tool for predicting sites of metabolism (SoMs). In this work, we focus on the prediction of the chemical structures of metabolites, in particular metabolites of xenobiotics. To this end, we have developed a new tool, GLORY, which combines SoM prediction with FAME 2 and a new collection of rules for metabolic reactions mediated by the cytochrome P450 enzyme family. GLORY has two modes: MaxEfficiency and MaxCoverage. For MaxEfficiency mode, the use of predicted SoMs to restrict the locations in the molecule at which the reaction rules could be applied was explored. For MaxCoverage mode, the predicted SoM probabilities were instead used to develop a new scoring approach for the predicted metabolites. With this scoring approach, GLORY achieves a recall of 0.83 and can predict at least one known metabolite within the top three ranked positions for 76% of the molecules of a new, manually curated test set. GLORY is freely available as a web server at https://acm.zbh.uni-hamburg.de/glory/, and the datasets and reaction rules are provided in the Supplementary Material.

Keywords: cytochrome P450; metabolism prediction; metabolite structure prediction; metabolites; rule-based approach; sites of metabolism; xenobiotic metabolism.

Figure 1

Workflow for GLORY indicating the difference between MaxCoverage mode and MaxEfficiency mode.

Figure 2

Illustration of the determination of the maximum SoM probability of all heavy atoms in the parent molecule that are matched by the reaction rule, using the N-dealkylation reaction rule (common reaction type; factor F = 5) as an example. This maximum probability is used to calculate the priority score of the product.

Figure 3

Receiver operating characteristic (ROC) curves over the entire reference dataset of 848 compounds with 1,588 known metabolites, with varying values of the factor used for common reaction types when calculating the priority score for each metabolite. Note that a factor of 1 means that only the SoM probability (i.e., the maximum SoM probability for all atoms that are matched by the SMIRKS) affects the priority score of the predicted metabolite, regardless of the reaction type. Note also that a ROC curve can be calculated despite there being no “true negative” predictions overall (all predicted metabolites are “positive” predictions). To generate the ROC curve, the false positive rate (FPR) is calculated at each score threshold. At each point, predictions with scores below the threshold are considered “negative” predictions and predictions with scores above the threshold are considered “positive” predictions. Hence the number of “true negative” predictions and therefore the FPR can be calculated for each point of the ROC curve.

Figure 4

Precision (portion of predictions that are true positives) and recall (portion of known metabolites that are predicted) vary according to the cutoff for FAME 2's predicted SoM probability. A SoM probability cutoff of 0.4 corresponds to the decision threshold used in FAME 2. The SoM probability cutoff chosen for the MaxEfficiency mode of GLORY was 0.2.

Figure 5

Histograms of the recovery rate of known metabolites broken down by parent compound: (A) GLORY in MaxCoverage mode, (B) GLORY in MaxEfficiency mode, (C) SyGMa, (D) BioTransformer. For example, a recovery rate of 0.5 indicates that for x% of all parent molecules, at least half of all recorded metabolites from the test dataset were predicted.

Figure 6

Histograms of the number of putative false positive predictions: (A) GLORY in MaxCoverage mode, (B) GLORY in MaxEfficiency mode, (C) SyGMa, (D) BioTransformer. These histograms use right-closed intervals.

Figure 7

ROC curves over the entire test dataset comparing the (A) scoring and (B) ranking approaches of SyGMa to GLORY's MaxCoverage mode. For a better comparison of the two methods, false negatives were included in the ROC curve by assigning those data points a score of 0 or rank of 1,000, as applicable.