Fig. 2

Previously published CRC subtypes do not separate samples’ transcriptional profiles. Average silhouette widths between the previously reported CMS [10] subtypes provide no evidence for substantial clustering structure. Silhouette widths for separation between CMS subtypes are calculated within each of the 18 datasets used in our study, either for all samples or only for those confidently labeled by the CMS classifier as provided in [10]. Distribution of samples’ silhouette widths is represented with box plots, with diamonds marking the average. Studies are separated according to training and validation sets, and then ranked based on their average silhouette widths. Datasets also used in the CMS paper are marked in red. The reference levels of clustering (horizontal gray dashed lines) are the same as in [25]. These results are not sensitive to dissimilarity measures (Additional file 1: Figure S1B)