Skip to main content
Fig. 4 | Genome Biology

Fig. 4

From: Continuity of transcriptomes among colorectal cancer subtypes based on meta-analysis

Fig. 4

Continuous subtype scores consistently reproduce CMS subtypes, but provide additional information in characterizing molecular/histopathological/clinical correlates. a Subtype- and study-specific mean PCSS1 and PCSS2 scores indicate “quadrants” in the distribution of the continuous scores that correspond to CMS1–4 subtypes, with unlabeled samples clustered at the origin. Each point indicates the average PCSS1 and PCSS2 value of samples classified as a particular CMS subtype in one dataset, with error bars representing standard deviation. b CMS4 subtype has the worst DFS outcome in all samples where survival information is available, agreeing with results in [10], but stratification of CMS4 samples with respect to continuous scores reveals an even more highly at-risk subgroup at the extreme end of PCSS1/PCSS2 distributions. Individual hazard ratios for each study are included in Additional file 1: Figure S9C). Continuous scores are more closely associated with molecular and clinical/pathological variables than discrete subtypes. Molecular, histopathological, and clinical variables were regressed on subtypes and scores as covariates. LRTs were used to compare the full model, containing both subtype and score as predictors, to a simplified model containing only subtype (left) or score (right) as predictor. Test results for different datasets (p values) are represented by points in the box plots. A p value near 1 (−log-10 p value near 0) suggests that no additional information is provided by the full model, whereas a small p value suggests that the full model provides additional information for predicting molecular/clinical variables. The more significant p values for models using only discrete subtypes (left) vs continuous scores (right) suggest that discrete subtypes alone lack information provided by the full model; conversely, log-10 p values near zero for scores (right) suggest that continuous scores outperform discrete subtypes in characterizing the molecular and clinical/pathological variables

Back to article page