Poster 54: Practical Application of Selected Computational Methods for Prediction of Xenobiotic Metabolism: Rational Strategies for UsePrzemyslaw Piechota1, Mark T. D. Cronin1, Mark Hewitt1, Judith C. Madden1
|1School of Pharmacy and Biomolecular Sciences, Liverpool John Moores University, Byrom Street, Liverpool L3 3AF, England|
|Prediction of xenobiotic metabolism is a research priority in many areas including pharmaceutical, cosmetic, food safety and environmental studies. A range of software has been developed capable of predicting potential metabolites and / or likely sites of metabolism (SOM). The performance of a selection of such software was investigated in this study. Meteor (Lhasa Limited, Leeds), SMARTCyp (http://www.farma.ku.dk/smartcyp/index.php) and MetaPrint2D-React (http://www-metaprint2d.ch.cam.ac.uk/metaprint2d-react) were used. The algorithms were assessed using two datasets; one a homogenous dataset of 28 Non-Steroidal Anti-Inflammatory Drugs (NSAIDs) and paracetamol (DS1); the second a diverse dataset of 30 top-selling drugs (DS2). Known metabolites for the 59 drugs were collated from the literature. The known metabolites were compared to those predicted by the software. The prediction of metabolites for the diverse dataset (DS2) was better (i.e. more of the known metabolites were correctly predicted by the software) than for the more homogenous DS1, indicating that some areas of chemical space may be better represented than others in the data used to develop and train the models. The study also identified compounds for which none of the packages could predict metabolites, again indicating areas of chemical space where more information is needed. |
Further investigation was carried out into pragmatic approaches to using metabolism prediction software more efficiently. The Meteor software package correctly predicted the majority of metabolites (in the datasets studied here) although this was achieved at a cost of low precision. Different settings for absolute reasoning (probable, plausible, equivocal) and relative reasoning (rel1, rel2, rel3) were explored in order to address this issue. The number of predictions present in the output was limited by setting nine cut-off points for the output (i.e. 5, 10, 15, 20, 25, 30, 35, 40 and 400 metabolites). It was found that use of cut-off values above 25 and 30, for DS1 and DS2 respectively, the number of correct predictions did not change significantly. Therefore, using cut-off values instead of restrictive reasoning settings can lead to a reduction in the output with little loss of sensitivity. It indicates that in terms of drug development a more pragmatic approach to predicting metabolites using Meteor may be to use low-filter reasoning settings (such as equivocal / rel2 or rel3) but to limit the number of predicted metabolites investigated to 30. This would offer significant savings in time and effort in predicting likely metabolites that can subsequently be processed more efficiently for toxicity assessment.
A combination of freely available packages was also used to predict metabolites. The software included: SMARTCyp (which predicts SOM) and MetaPrint2D-React (which predicts SOM and structures of metabolites). The rationale behind selecting a combination of these two approaches is that the SMARTCyp method, in contrast to other software such as MetaPrint2D, does not depend on historical reaction data but uses an empirical approach (pre-calculated energies for a number of sub-fragments). One of the benefits of such a method is that it could be applied with greater confidence to predict SOM for molecules that do not share substantial structural similarity with compounds found in historical databases. Therefore, such predictions would be less biased (towards already existing data) providing that the array of sub-fragments is large enough to cover a variety of structures. However, SMARTCyp does not provide metabolites’ structures, which may be required by some users. Hence, MetaPrint2D-React was used to furnish the structures. A two-step workflow was devised in which SMARTCyp is used to predict likely metabolic hot spots and MetaPrint2D-React is used to provide possible biotransformations (metabolite structures). The workflow was applied to DS2, the metabolites of interest being those known to be products of CYP3A4 (39 primary metabolites for 17 drugs) and CYP2D6 (11 primary metabolites for 6 drugs). Analysis of the predictions obtained using this combination of methods indicates that the CYP2D6 model in SMARTCyp performed better than the CYP3A4 model. The CYP2D6 model predicted 91% of metabolites correctly using the five top-ranking SOM and predicted 73% of metabolites when using only the three top-ranking SOM. For CYP3A4 the model predicted 56% of metabolites correctly using the five top-ranking SOM and 44% correctly using the three top-ranking SOM. In all cases but one, MetaPrint2D-React was able to list the correct structures of metabolites. The described approach could be further developed (extending beyond primary metabolites) and implemented to produce a metabolic tree, as such it could provide an alternative method for prediction of metabolic transformations.
Funding provided by the eTOX project, grant agreement number 115002 under the Innovative Medicines Initiative Joint Undertaking (IMI-JU) is gratefully acknowledged.