ISMARA: automated modeling of genomic signals as a democracy of regulatory motifs

Outline of the Integrated System for Motif Activity Response Analysis. (A) ISMARA starts from a curated genome-wide collection of promoters and their associated transcripts. Using a comparative genomic Bayesian methodology (Arnold et al. 2012a), transcription factor binding sites (TFBSs) for ∼200 regulatory motifs are predicted in proximal promoters. Similarly, miRNA target sites for ∼100 seed families are annotated in the 3′ UTRs of transcripts associated with each promoter (Friedman et al. 2009). (B) Users provide measurements of gene expression (microarray, RNA-seq) or chromatin state (ChIP-seq). The raw data are processed automatically, and a signal is calculated for each promoter in each sample. For ChIP-seq data, the signal is calculated from the read density in a region around the transcription start. For gene expression data, the signal is calculated from read densities across the associated transcripts (RNA-seq) or intensities of associated probes (microarray). (C) The site predictions and measured signals are summarized in two large matrices. The components N_pm of matrix N contain the total number of sites for motif m (TF or miRNA) associated with promoter p. The components E_ps of matrix E contain the signal associated with promoter p in sample s. (D) The linear MARA model is used to explain the signal levels E_ps in terms of bindings sites N_pm and unknown motif activities A_ms, which are inferred by the model. The constants c_p and correspond to basal levels for each promoter and sample, respectively. (E) As output, ISMARA provides the inferred motif activity profiles A_ms of all motifs across the samples, s, sorted by the significance of the motifs. A sorted list of all predicted target promoters is provided for each motif, together with the network of known interactions between these targets (provided by the String database, http://string-db.org/) and a list of Gene Ontology categories that are enriched among the predicted targets. Finally, for each motif, a local network of predicted direct regulatory interactions with other regulators is provided.