New chemical development and hazard assessments benefit from accurate predictions of mammalian and microbial metabolites. Fourteen biotransformation libraries encoded in eight software packages that predict metabolite structures were assessed for their sensitivity (proportion of reported metabolites that were predicted) and selectivity (proportion of predicted metabolites that were reported) toward reported mammalian and microbial metabolites for 10 agrochemicals. No library averaged greater than 58% sensitivity or 69% selectivity. With an increasing number of predicted generations, sensitivity increased and selectivity decreased. To test whether consensus predictions performed better than individual libraries, metabolites predicted by at least two of the top three, four, and five libraries in ease of use and sensitivity were considered. Almost every consensus group had greater sensitivity and selectivity than the constituent libraries. Gaps in the predictions were identified by comparing the reaction types that lead to each reported metabolite and the frequency with which each metabolite was predicted.

Assessing the accuracy of software predictions of mammalian and microbial metabolites

250th National Meeting (2015)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.