P System Model Optimisation by Means of Evolutionary Based Search Algorithms
Executable Biology, also called Algorithmic Systems Biology, uses rigorous concepts from computer science and mathematics to build computational models of biological entities. P systems are emerging as one of the key modelling frameworks within Executable Biology. In this paper, we address the continuous backward problem: given a P system model structure and a target phenotype (i.e. an intended biological behaviour), one is tasked with finding the (near) optimal parameters for the model that would make the P system model produce the target behaviour as closely as possible. We test several real-valued parameter optimisation algorithms on this problem. More specifically, using four different test cases of increasing complexity, we perform experiments with four evolutionary algorithms, and one variable neighbourhood search method combining three other evolutionary algorithms. The results show that, when there are few parameters to optimise, a genetic and two differential evolution based algorithms are robust optimisers attaining the best results. However, when the number of parameters increases, the variable neighbourhood search approach performs better.
In press in Proceedings of the Genetic and Evolutionary Computation Conference