Comparing separate clustering and iCluster performance using simulation.

Simulation scenario 1 consists of a pair of matched data sets of 200 features (20 of which are relevant to clustering and 180 are noisy features) in 100 samples belonging to two distinct clusters. Scenario 2 represents an extremely sparse data structure with only 2 cluster-associated features and 198 noisy features. RI is the resampling-based cluster reproducibility criterion and ranges between 0 and 1. A value close to 1 indicates perfect cluster reproducibility, and a value close to 0 indicates poor reproducibility. Separate K-means has two sets of numbers associated with each criterion because of separate model fits. The numbers are similar and therefore averaged in the table. The number in parentheses is the standard deviation over 50 simulations.