Figure 1.
This figure refers to the definition of a densely connected bicluster (see Methods section, definition 1) referring to the parameters (density) and
(co-expression constraints).
The input for the core algorithm is the interaction network of the organism (here, as a toy example, genes A,B,C,D,E,F,G,H,I,J and K) together with gene expression dataset containing (logarithmic) fold changes of genes across a set of experimental conditions (here: the table below the interaction network). On the right, we display the set of densely connected biclusters which refer to the datasets on the left. The densely connected biclusters contain at least times the amount of possible edges and its genes are co-expressed in at least
different experimental conditions (
) with a difference of at most
(
).
Table 1.
Benchmarking competition yeast.
Table 2.
Benchmarking competition human.
Figure 2.
Runtimes of our algorithm for varying, biologically relevant choices of the parameters involved in our framework.
The most important observation is that we have runtimes of at most a few minutes for all choices of (density).
Table 3.
Statistics on overlapping module pairs supporting different functionalities (OMPSDF).
Figure 3.
Two real case examples of a Yeast (left) and a Human (right) module as inferred by application of DECOB and further filtering by GO terms of specific interest.
The Yeast module on the left was obtained by screening the output of DECOB for modules which are enriched with the GO term “Chromatin Assembly” (GO:0006333). The Human module on the left was obtained by screening the output of DECOB for modules which are enriched with the GO term “Wnt Receptor Signaling Pathway through Beta-Catenin” (GO:0060070).
Figure 4.
Illustration of the DECOB algorithm on a simplified example consisting of six genes and three gene expression conditions.
DECOB constraints are specified by: (density),
(maximum difference in expression) and
(number of expression conditions). The algorithmic strategy is to traverse the lattice of all subnetworks in a breadth-first fashion. Any subnetwork which is not a densely connected biclusters can be discarded due to that every densely connected bicluster necessarily has a densely connected bicluster as a parent ( = subnetwork contained in the original one, see definitions 1 and 3 and the surrounding discussions). For esthetical reasons, we have omitted B-C-D-E although, as a child of the densely connected bicluster B-C-D, it is also examined. B-C-D-E, just as A-B-D-E will be discarded since it violates the density constraint.