Data-based optimal control of gene regulatory networks

Abstract: In this presentation, a problem of decision making in gene regulatory networks is considered. Such decisions are made based on the measurements of the system's behaviour, which in gene regulatory networks is typically accompanied by a large degree of variability and uncertainty. Hence, in order to make a decision, efficient and novel computational methods are required. In the capacity of such a method, we propose to adapt fitted Q iteration from the reinforcement learning discipline. In order to perform its computations, the fitted Q iteration algorithm requires input (e.g., a schedule of light pulses, which affect the system) and output (e.g., protein concentrations) data sets of a gene regulatory network. This data set can either be collected from wet-lab experiments or be created by computer simulations of the dynamical model of the system. The output of the algorithm is a feedback control policy, i.e., a rule specifying which control action should be taken based on the measured system's state. This technique is applicable to a wide range of biological systems, due to its inherent ability to deal with highly stochastic system dynamics. In order to illustrate the performance of the approach, two problems are considered: regulation of the toggle switch system, where the objective is to drive the system to a prescribed point, and reference tracking in the generalised repressilator system, where the objective is to make the system follow a prescribed trajectory or a reference.