Better than random? The chemotype enrichment problem

Chemotype enrichment is increasingly recognized as an important measure of virtual screening performance. However, little attention has been paid to producing metrics which can quantify chemotype retrieval. Here, we examine two different protocols for analyzing chemotype retrieval: "cluster averaging", where the contribution of each active to the scoring metric is proportional to the number of other actives with the same chemotype, and "first found", where only the first active for a given chemotype contributes to the score. We demonstrate that this latter analysis, common in the qualitative analysis used in the current literature, has important drawbacks when combined with quantitative metrics.