Fig. 1

Overview of survClust. a A simulated data example, consisting of features that define 3 patient subtypes without direct association with survival (shaded in red), features that define 3 patient subtypes with distinct survival outcome (shaded in blue), and random features generated from Gaussian noise (gray). b Euclidean distance matrix demonstrating patient-level pairwise similarity, with darker blue shade representative of higher similarity. Color panels above the distance matrix show the three-class solution obtained by unsupervised algorithm via k-means and the concordance between the simulated 3 survival subtypes (the truth). Kaplan-Meier curves for the 3 unsupervised subtypes show no distinction in survival outcome. c survClust employs a patient outcome-weighted distance matrix to identify the desired subtypes with distinct Kaplan-Meier curves. d survClust allows integrative analysis of multiple data modalities to identify survival-associated molecular subtypes