Session 4: In silico metabolism

A hybrid expert system-machine learning approach to xenobiotic metabolite ranking
A hybrid expert system-machine learning approach to xenobiotic metabolite ranking

Carol Marchant


Lhasa Limited


Research Leader


In silico structure-activity relationship models for the assessment of toxicity or associated events often include implicit consideration of metabolism.

However, metabolic relationships may be missed if, for example, metabolic precursors are not sufficiently represented in the training data set or the assay system on which the model is based is not fully metabolically competent.

In silico prediction of the metabolism of a chemical of interest can therefore add value to its toxicological assessment. Since it is not feasible to assess the toxic potential of all possible metabolites it is important that the most likely ones can be prioritised by ranking methods.

The Meteor Nexus expert system has traditionally made use of a qualitative model of absolute and relative reasoning to rank metabolites.

An alternative approach which relies on machine learning from the results of experimental metabolite identification studies for structurally similar substrates will be described.

The method generates quantitative likelihoods which can be more easily ranked than qualitative terms and integration of these values with in silico assessments of toxicity allow the easy identification of metabolites that are both likely to form and for which there is high confidence in the prediction of toxicity.

Currently the methodology makes use of a collection of metabolic pathways which have been harvested from public domain sources. The potential exists for further optimisation through the expanded chemical space offered by the private or shared use of proprietary metabolism data subject to investment in their suitable electronic storage.