Gene selection for cancer types classification using novel hybrid metaheuristics approach

Abstract

Abstract With the advancement of microarray technology, gene expression profiling has shown remarkable effort to predict the different types of malignancy and their subtypes. In microarrays, predicting highly discriminative genes is a challenging task and existing hybrid methods fail to deal with efficiently. To mitigate the curse of dimensionality problem and to improve the interpretability of discriminative genes, in this study, we developed a new hybrid wrapper approach which integrates the characteristics of teaching learning-based algorithm (TLBO) and gravitational search algorithm (GSA), called TLBOGSA. A new encoding strategy is also integrated into TLBOGSA to transmute the continuous search space to binary search space and form binary TLBOGSA. In the proposed method, firstly, minimum redundancy maximum relevance (mRMR) feature selection is employed to select relevant genes from the gene expression datasets. Then, wrapper method is applied to select the informative genes from the reduced data produced by mRMR. To improve the search capability during the evolution process, we have incorporated the gravitational search mechanism in the teaching phase. The proposed method uses naive bayes classifier as a fitness function to select the extremely judicious genes which can help to classify cancer accurately. The efficiency of proposed method is tested on ten biological datasets and compared with state-of-art computational intelligence approaches for tumor prediction. Experimental results and statistical analysis demonstrate that proposed method is significantly outperforms existing metaheuristic approaches regarding convergence rate, classification accuracy and optimal number of feature sets. The proposed method reaches above 98% classification accuracy in six datasets and maximum accuracy is achieved as 99.62% in DLBCL dataset.