Study On The Function Generalization And Intervention In Gene Regulatory Network

Boolean network has been a major model to study gene regulatory networks. Lots of work have been focused on inferring networks from time-series data and designing potential intervention policies. However, one important problem still remains unsolved, that is the generalization of Boolean function. In general, the inference algorithms always assume a random Boolean value for the unobserved states.As many theoretical and experimental results support that gene regulatory networks lie between the boundary of ordered and disordered regimes, we studied three generalization methods: the majority rule, bias-based and mutual information-based methods in the first part of this paper. Results both on simulation networks and melanoma network show that reasonable generalization can improve both the steady-state distribution distance and the sensitivity error. And among the three methods, the mutual information-based method performs better than the other two.Gene regulatory network is ultimately to serve network intervention. As is known, given the complete network, the optimal policy performs better than the suboptimal policy. However, this may not hold if we intervene in a system based on a control policy derived from imprecise inferred networks, especially in the small-sample scenario. In the latter part of this paper, we compare the performance of the unconstrained(UC) policy with the mean-first-passage-time(MFPT) policy with regard to two aspects: the quality of the determined control gene and the effectiveness of the policy. The results reveal that: in the inferred networks, the MFPT policy working better in the small-sample scenario and the UC policy working better only for the large-sample scenario. Additionally,using the relatively complex model(N>1) is beneficial for the intervention process, especially for the sensitive UC policy.