Figure 2.

Illustration of different components of CMRF and connectivity between them. (A) obtains an initial estimates of state variables using Student's t test. (B) estimates parameters in a way that maximizes data likelihood. (C) estimates parameters in order to maximize prior density. Both (B) and (C) use a global optimization technique called Differential Evolution. (D) employs Iterated Conditional Modes to maximize the pseudo-likelihood. (B), (C) and (D) consist of an alternating optimization technique. These three steps (B), (C) and (D) are repeated till the algorithm meets a criteria for completion. Finally, once the optimization is complete, the DR genes are sorted in decreasing order of their likelihood with respect to the metagene g0. The genes at the top of the list are declared PDR.

Bandyopadhyay et al. BMC Genomics 2012 13(Suppl 2):S2   doi:10.1186/1471-2164-13-S2-S2