Evolutionary computation and swarm intelligence for the inference of gene regulatory networks

Abstract

The inference of gene regulatory networks (GRNs) from expression profiles is still an important challenge in bioinformatics research. The main difficulty of this problem is associated to the huge number of genes and the small number of samples available, as well as the intrinsic noise in the data acquisition process. In this context, this paper presents a feature selection approach to the identification of GRNs using optimisation strategies from evolutionary computation and swarm intelligence. As a case-study we used an artificial gene network (AGN) based on the scale-free topology. This AGN has 1,000 genes and was simulated with 500 temporal expression samples. The methods compared were: differential evolution (DE), bat algorithm (BAT) and artificial bee colony (ABC) algorithms. All algorithms used their standard control parameters and the same criterion function: the mean conditional entropy (MCE). This is an information theory measure, commonly adopted for various feature selection problems in the pattern recognition research field. The results showed that DE algorithm leaded to the best results than BAT and ABC in all comparisons, and the inferred network was more similar to the original network.