Abstract
The problem decomposition strategy is a very efficient technique for the inference of S-system models of genetic networks. This strategy defines the inference of a genetic network consisting of N genes as N subproblems, each of which is a 2(N+1)-dimensional function optimization problem. Genetic networks made up of dozens genes can be analyzed with this strategy, though the computational cost in doing so remains quite high. In this study, we attempt to infer S-system models more efficiently by further dividing each 2(N+1)-dimensional subproblem into one (N+2)-dimensional problem and one (N+1)-dimensional problem. The subproblems are divided using the genetic network inference method based on linear programming machines (LPMs). Next, we propose a new method for estimating the S-system parameters by alternately solving the two divided problems. According to our experimental results, the proposed approach requires less than one-third of the time required by the original problem decomposition approach. Finally, we apply our approach to actual expression data from the bacterial SOS DNA repair system.
