Fig. 1

Similarity scores performance a Spectral similarity S1 vs. Intra-cluster morphological similarity S2 scatter plot. Each point represents a potential cluster classified by the algorithm. All clusters shown in Table 2 are evaluated for all 14 datasets presented in Table 1. Blue points represent the “positive class” (should be present in the sample) while the red points correspond to the negative class (should not be present in the sample). Most “positive class” points are located in the top right corner well separated from the negative class points. This indicates proper classification power. b Precision and recall (PR) curve computed according to Davis et al. 2006 [30]. c Similarity score S1·S2 vs. Cluster number. Clusters are arranged in decreasing order of mean similarity score. A clear gap between an S of 0.5 and 0.7 separates the “positive class” from the negative class. Refer to Additional file 1: Table S1 for a mapping of cluster numbers to cluster chemical formula