Gene expression profiling by estimating parameters of gene regulatory network using simulated annealing: A comparative study

Gene regulation is either an intra-cellular, inter-cellular, intra-tissue or inter-tissue biochemical phenomenon in an organism where a few genes may regulate the expression(s) of any other gene(s), even the expression of itself. The regulation is performed through proteins, metabolites and other genetic spin-offs resulting from the change in environment that genes experience in the cellular context. The gene regulatory network which originates from the regulation process is a potential source from which different physiological, behavioral, medicinal and disease-related issues of an organism can be uncovered. Computational inference of the network is a well-known bioinformatics task. Easy availability of time series gene expression data has made the work easier. But this data suffers from the curse of dimensionality as columns (time points) are few in number in comparison with rows (genes). Methods which are proposed here take the microarray time series gene expression data as input and simulate a time series of larger number of rows with regular small intervals. The parameters of the gene regulatory network are estimated using three variants of Simulated Annealing, viz. Basic Simulated Annealing (BSA), Tabu Simulated Annealing (TSA) and Greedy Simulated Annealing (GSA). During the estimation of parameters, the main focus is on minimizing the cost between actual and simulated time series in successive iterations. The final parameter set is used to produce the simulated time series, each row of which is the expression profile of a gene. With an available synthetic data set, original expression profiles are compared to the expression profiles produced by three different methods. The simulated profiles show close correspondence to the original ones. GSA shows the closest correspondence and TSA proves to be the most efficient in terms of time and number of iterations. The simulated time series may be used for GRN reconstruction or other problems.