CLIC, a tool for expanding biological pathways based on co-expression across thousands of datasets
Fig 1
CLIC partitions an input Query gene set into co-expressed modules (CEMs), assigns weight to each dataset according to the intra-correlation of each module relative to background, and then predicts additional genes co-expressed with each CEM in high-weight datasets. CLIC inputs a compendium of D microarray data sets (e.g. from GEO) and an input Query gene set. In the Partition step, input genes are partitioned into distinct CEMs (in this example, CEM 1 in red, CEM 2 in orange), using a Bayesian partition model to simultaneously infer the number of CEMs and assign weights to datasets. Dataset weights quantify the significance of each intra-CEM correlation compared to the background distribution of correlation in each dataset (gray density curves). Genes from the input set that are not assigned to any CEM are assigned to a “Null” cluster. In the Expansion step, each CEM is expanded by identifying additional genes that show higher co-expression with the CEM genes compared to the gene-specific background distribution, scored by the log-likelihood ratio (LLR).