Fig. 2

Overview of the CoMEt algorithm. First, we transform alteration data from different measurements into a binary alteration matrix A. Second, we use a Markov chain Monte Carlo (MCMC) algorithm to sample collections M, containing t sets of k alterations, in proportion to the weight Φ(M)^−α. Here we show a collection containing sets M and M ^′ with three and two alterations, respectively. We identify all collections whose weight exceeds the maximum observed in randomly permuted datasets. We summarize the alterations in these significant collections with a marginal probability graph, whose edge weights indicate the fraction of significant collections with the corresponding pair of alterations. Finally, we remove low-weight edges in the graph, obtaining the output modules