Identifying Gene Regulatory Network as Differential Equation by Genetic Programming

This paper proposes an evolutionary method of identifying the gene regulatory network represented as a differential equation system. As the technology in DNA micro arrays has developed, large quantities of gene’s expression data are becoming more available. As a result, it is essential to get information as to the gene regulatory network from the observed data of gene’s expression. Among many proposed models to describe a gene network, we have chosen the differential equation system since it can represent complex relations among components. In the previous studies [1], the form of the differential equation is being fixed during the learning so that the ultimate goal of the identification is to optimize parameters, i.e., coefficients, in the fixed equation. On the other hand, for the sake of the flexibility of the model, we allow an arbitrary form of functions in the right-hand side of the differential equation (eq. (1)). dXi/dt = fi(X1, X2, . . . , Xn) (1)