Figure 2

Simulated data. A) & C) Predicted pathway activity levels (pal) (y-axis) against sample-ID (x-axis) for specific runs of the simulated data. Pathway activity levels were estimated separately using the three algorithms (i) unpruned average metric (UPR-AV), (ii) pruned average metric (PR-AV) and (iii) pruned weighted average metric (DART). The variational Bayesian clustering approach was used to infer the clusters in these pathway activity level profiles. Black denotes the inferred cluster with lowest mean pathway activity level, red denotes samples assigned to higher level clusters. Downward pointing triangles denote the samples with no true pathway activity (samples 1-40), upward pointing triangles denote the samples with true pathway activity (albeit variable levels) (samples 41-100). The proportion of correctly assigned samples is given (Accuracy). A) refers to SimSet2 and C) refers to SimSet1. B) & D) Comparison of classification accuracies (y-axis) of the three algorithms (x-axis) over 100 simulations. B) SimSet2. Wilcoxon-test P-values given are between UPR-AV and PR-AV, and between PR-AV and DART. C) SimSet1. Wilcoxon-test P-value is between PR-AV and DART.