Figure 1

Overview of our method. In Step 1, we generate all possible putative transcriptional modules (PTMs) from ChIP-chip data as described in Methods section. Here we show a toy example of ChIP-chip binary data matrix and corresponding 4 PTMs, M1 to M4. Each module contains a set of transcription factors (triangles) and a set of target genes (circles) connected by links between all of them (bicliques; see Eq. (1)). The genes in M2 and M3 are numbered for an illustration purpose below. In Step 2, we identify coherent modules among all PTMs using gene expression and functional annotation data. Only those PTMs which satisfy Eq. (5) are selected as coherent modules. In the given example in the figure, M1 and M4 are meant to be non-coherent and hence discarded altogether. Colours of genes in both coherent modules, M2 and M3, symbolize different functions. 'Blue' and 'red' functions are meant to be coherent (enriched) in the respective modules. Notice that the 'red' function is coherent in both modules. The fictitious 'M'-shaped expression profiles are also shown to be coherent as well in both modules. The red profile belongs to gene 5 which is annotated to the coherent 'red' function. Finally, in Step 3, we identify those genes which appear in multiple coherent modules with any common coherent functions. This is illustrated in Step2 by the gene 5 which belong to both coherent modules. The gene is annotated to the common coherent 'red' function in both modules. The union of regulators in M2 and M3 is predicted to functionally regulate the gene in this illustration. We term such genes 'coherent linker genes'. Notice that gene 6 belonging to both modules is not a coherent linker gene because its annotated 'yellow' function is not coherent in the modules. See the text for more details.