Gene expression profiling by estimating parameters of gene regulatory network using meta-heuristics: A comparative study

Abstract

Genes are present in the nucleus of every cell in an organism. Genes, metabolites, proteins and other by-products of cellular activity form a signaling pathway or network which is called a Gene Regulatory Network. Computational reconstruction of the network may uncover potential genetic causes of diseases and may aid drug detection. Advancements in biotechnology and image processing tools have made time series gene expression data available to researchers of computational biology. Reconstruction of Gene Regulatory Network has found a new direction with the availability of this data. After being processed by different statistical methods, the time series data may be considered as a matrix with each row representing a gene and each column representing a time point. The data suffers from an insufficiency of number of columns in relation to number of rows. This makes the reconstruction process more tedious. The problem is known as Curse of Dimensionality problem. The methods which are described here take processed microarray gene expression data as the input and produce the simulated gene expression time series with larger number of columns having regular small intervals. Gene Regulatory Network is reconstructed in the framework of Recurrent Neural Network. The parameters of the network are iteratively optimized using efficient local search optimization algorithms, namely two variants of Simulated Annealing and Tabu Search. The optimized parameters are used for the comparative study between the three methods in producing the time behavior or expression profiles of the genes. For almost all genes, the simulated profiles closely correspond to the original profiles.