On the performance of particle swarm optimization for parameterizing kinetic models of cellular networks

Recent advances in high-throughput technologies and an increased knowledge of biochemical systems have enabled the reconstruction of cellular regulatory networks. While these reconstructions often consist of the connectivity patterns among the network´s components, an important task in this area is to derive values for the kinetic parameters of the system so that its dynamics can be elucidated and testable hypotheses can be generated. The problem here is to reverse engineer the model (in our case, values of kinetic parameters) given the data that consists of a network connectivity and time-series data. Particle swarm optimization (PSO) has been recently used for this reverse engineering task. In this paper, we introduce three scoring metrics for assessing the optimality of the solution found by PSO. Using biological data sets, we study the performance of the PSO framework under the three scoring metrics. Our results show that PSO achieves very good parameterization of the cellular networks we consider, regardless of the complexity of the dynamics patterns generated by the underlying kinetics. Given the flexibility of the PSO framework, as well as its natural amenability to parallelization and general high-performance implementations and execution, our results indicate PSO may be a good candidate for parameterizing cellular networks.