Compromised intervention policies for phenotype alteration

We take a Markovian approach to modeling gene regulatory networks and assume that phenotypes are characterized by the steady-state probability distribution of such networks. We desire intervention policies that maximally shift the probability mass from undesirable states to desirable ones. In doing so, we might also be concerned about the steady-state mass of some “ambiguous” states, which are not directly related to the pathology of interest but could be associated with some anticipated risks. We propose a direct formulation of this constrained optimization problem, rather than assuming a subjective cost function, and provide optimal intervention policies. Within this framework, we investigate the performance of “compromised” policies, these being policies for which we accept some increase of the ambiguous mass to achieve more decrease in the undesirable mass.