Gene Regulatory Networks Full Observable Cbased on Batch Reinforcement Learning: An Improved Policy

The principal objective of controlling gene regulatory networks (GRNs) as dynamic biochemical systems is to find effective treatment methods, also studying gene impact on diseases and their development. The control problem aims at the intervention of appropriate strategy to avoid diseased states through some input genes, called control genes. One of the critical challenges in present external control techniques is the increase of time and memory complexity exponentially regarding the number of genes over the network. On the other hand, the methods cannot be utilised even for a few dozens of genes. An approach to solve this problem is to control the network fully observability using directly obtained experience tuples from network states. The idea uses an approximate strategy based on batch reinforcement learning and least squares linear regression to find a sequence of actions. In the current study, we propose an improved strategy based on the previous method to control fully observable GRNs. Our idea can provide a control policy for thousands of genes. Results demonstrate that our methodology is more effective than the previous study and provides a better functionality related to inspected criteria.