CLIC (Clustering by Inferred Co-expression)

CLIC clusters genes by inferred co-expression to identify co-expressed pathway modules (CEMs), predict novel pathway members, and determine the transcriptional contexts in which these modules operate.

Key Features:

• Input: Accepts a pathway consisting of two or more genes as the input gene set.
• Bayesian partition model: Uses a Bayesian partition model to partition the input gene set into coherent co-expressed modules (CEMs) and assigns each CEM a posterior probability per dataset indicating its likelihood of representing a true biological module.
• Module expansion: Expands identified CEMs by scanning the transcriptome to identify additional co-expressed genes.
• Integrated LLR scoring: Scores candidate gene additions using an integrated log-likelihood ratio (LLR) weighted according to each dataset's contribution.
• Contextual learning: Automatically identifies the specific conditions or datasets in which each CEM operates.
• Dataset compendium: Applies analyses across a compendium of transcriptional profiling data including 1774 mouse microarray datasets (comprising 28,628 microarrays) and 1887 human microarray datasets (encompassing 45,158 microarrays).

Scientific Applications:

• Pathway component discovery: Predicts novel members of biological pathways from co-expression patterns across large transcriptional compendia.
• Context-specific pathway activity: Identifies datasets and conditions where pathway modules are active, aiding interpretation of pathway relevance across biological states.
• Validated example: Predicted a functional connection between protein C7orf55 (FMC1) and the mitochondrial ATP synthase complex that was subsequently validated experimentally.

Methodology:

Input a predefined gene set from a pathway; apply a Bayesian partition model to identify coherent co-expressed modules (CEMs) and assign posterior probabilities per dataset; expand modules by identifying additional co-expressed genes using an integrated LLR score weighted by dataset contributions; and determine operational contexts for each module across datasets.