Evaluation of Three State-of-the-Art Metabolite Prediction Software Packages (Meteor, MetaSite, and StarDrop) through Independent and Synergistic Use

The aim of this study was to evaluate three different metabolite prediction software packages (Meteor, MetaSite, and StarDrop) with respect to their ability to predict loci of metabolism and suggest relative proportions of metabolites. A chemically diverse test set of 22 compounds, for which in vivo human mass balance studies and metabolic schemes were available, was used as basis for the evaluation. Each software package was provided with structures of the parent compounds, and predicted metabolites were compared with experimentally determined human metabolites. The evaluation consisted of two parts. First, different settings within each software package were investigated and the software was evaluated using those settings determined to give the best prediction. Second, the three different packages were combined using the optimized settings to see whether a synergistic effect concerning the overall metabolism prediction could be established. The performance of the software was scored for both sensitivity and precision, taking into account the capabilities/limitations of the particular software. Varying results were obtained for the individual packages. Meteor showed a general tendency toward overprediction, and this led to a relatively low precision (∼35%) but high sensitivity (∼70%). MetaSite and StarDrop both exhibited a sensitivity and precision of ∼50%. By combining predictions obtained with the different packages, we found that increased precision can be obtained. We conclude that the state-of-the-art individual metabolite prediction software has many advantageous features but needs refinement to obtain acceptable prediction profiles. Synergistic use of different software packages could prove useful.