Multi-population adaptive genetic algorithm for selection of microarray biomarkers

Due to the fast development of DNA microarray technology, researchers have measured large-scale gene expression data in a single trial. However, the classification of microarray data is a challenging task for cancer detection and prevention since gene expression datasets are often exceeding tens of thousands of genes with a small number of tissues. In order to determine a robust gene signature from microarray data, many researchers have explored several gene selection methods for the prediction of cancer recurrence. However, there is no agreement on which gene selection technique produces optimal subsets of genes and avoids over-fitting and curse of dimensionality issues. This inspires us to design a new technique for gene selection, called hybrid multi-population adaptive genetic algorithm that can overlook the irrelevant genes and classify cancer accurately. The proposed hybrid algorithm comprises two phases. In the first phase, an ensemble gene selection method is used to filter the noisy and redundant genes in high-dimensional datasets by combining multi-layer and F-score approaches. Then, a wrapper is designed by multi-population adaptive genetic algorithm with support vector machine and naive Bayes classifiers as an objective function to identify the high-risk differential genes. The performance of the proposed approach is evaluated on ten microarray datasets of numerous tumor types. Furthermore, the comparative experiments demonstrate that proposed method outperforms the several state-of-the-art wrapper and filter methods in terms of classification accuracy with an optimal number of genes.