Functional Gene Networks: A Preliminary Study on a Modified Genetic Algorithm for Candidate Discovery in Large Microarray Datasets

This paper reports a preliminary study on the discovery of potential gene networks in large gene expression data by applying an island model genetic algorithm for optimizing search. Candidates for networks are defined as highly correlated gene expression subsets in test data derived from the Arabidopsis gene expression atlas - AtGenExpress. A study of algorithm parameters primarily focuses on fitness function calibration. Preliminary results show initial solution size to dominate the direction of chromosome evolution, while the fitness constant dictates the range of solution sizes which have such influence on the fitness function. Optimal fitness constants and initial solution size of chromosomes have been resolved to 1.001, and 30% of data size, respectively to minimize subset size influence, and bias Illness towards small highly correlated sets which are more biologically likely to form networks.