Datasets description.

Dataset: GEO or Array Express dataset identifier; Trained signatures: signatures which used that dataset as training sample, if any; Validation signatures: signatures which used that dataset as independent validation sample; Outcome: type of relapse used for that dataset; Minimum follow up: minimum follow up required for that dataset, when this info was available; Number of samples: number of samples contained in that dataset, showing good and bad prognosis’ separately between brackets; Clinical info: samples ranges of stage and microsatellite status when this information was available; Platform: datasets’ hybridization platform. * NA: the authors do not provide clinical information about MSI and/or stage. No info: Although authors provide clinical information in the paper, samples are not labelled with this information in GEO or ArrayExpress. a. Stage II and III samples from data sets GSE17536 and GSE17537 were jointly used to derive signature VL10, but the later did not include enough events at these stage subgroups. b. Signature JS09 was built with Duke’s A and D and validated with Duke’s B and C samples.