Bi-objective Genetic Algorithm with Rough Set Theory for Important Gene Selection in Disease Diagnosis

Abstract

Gene selection is a general phenomenon in the subject of bioinformatics where data mining and knowledge innovation plays a significant role in selecting an optimal set of genes regarding some useful evaluation functions. Gene selection based on single objective genetic algorithm may not provide the best solution due to varied characteristics of the datasets. If multiple objective functions are combined, an algorithm generally provides more important genes compared to the algorithm relying on a single criterion. Here, two criteria are united and a novel bi-objective genetic algorithm for gene selection is proposed, which effectively reduces the dimensionality of the huge volume gene dataset without sacrificing any meaningful information. The method uses nonlinear hybrid cellular automata for creating initial population and a novel jumping gene technique for mutation to maintain diversity in chromosomes of the population. It explores rough set theory and Kullback–Leibler divergence technique to define two fitness functions, which are conflicting in nature and are employed to approximate a Pareto-optimal solution sets. The best solutions of the proposed method provide the informative genes used for disease diagnosis. The replacement strategy for the creation of next generation population is based on the Pareto-optimal solution regarding both the fitness functions. The experimental results on the publicly obtainable microarray data express the importance of the identified genes and the effectiveness of the proposed informative gene selection mechanism.