Molecular Subtypes of Non-muscle Invasive Bladder Cancer Seth P. Lerner, A. Gordon Robertson Cancer Cell Volume 30, Issue 1, Pages 1-3 (July 2016) DOI: 10.1016/j.ccell.2016.06.012 Copyright © 2016 Terms and Conditions
Figure 1 Unsupervised Clustering for Protein-Coding Genes in 476 Tumors (A) Selecting a clustering input of 1,854 protein-coding genes using thresholds at the 90th-percentile variance and a mean abundance of 20 (red box). (B) Cumulative distributions (CDFs) of consensus membership, and delta CDF area plots from a ConsensusClusterPlus v1.36.0 run with Pearson correlations, PAM, 200 iterations, and a gene fraction of 0.95, for solutions with between two and ten clusters. Arrows mark 3-, 4-, and 5-cluster solutions. (C) Consensus membership heatmaps for 2-, 3-, 4-, and 5-cluster solutions. (D) For a 3-cluster solution, a heatmap of normalized gene abundance for 345 genes that were highly scored as differentially abundant. Below the heatmap is a profile of silhouette width calculated from the consensus memberships (Wcm), then a subset of covariates with Fisher exact p values. (E) Kaplan-Meier plot for progression-free survival (PFS), with a log-rank p value. (F) Relationships between cluster membership in 3- versus 4-cluster and 3- versus 5-cluster solutions. Each curve corresponds to one of the 476 samples. Silhouette width profiles, like that under the heatmap in (D), are placed vertically on either side. Cancer Cell 2016 30, 1-3DOI: (10.1016/j.ccell.2016.06.012) Copyright © 2016 Terms and Conditions