Artificial Bee Colony algorithm based on Dominance (ABCD) for a hybrid gene selection method

Abstract

In cancer research, it is important to classify tissue samples in different classes (normal, tumour, tumour type, etc.). Gene selection purpose is to find the minimum number of genes that can predict sample classes with efficacy. This work is focused on the gene selection problem by introducing a new hybrid method. This new method combines a first step of gene filtering with an optimization algorithm in a second step to find the best subset of genes for the classification task. The first step uses the Analytic Hierarchy Process, in which five ranking methods are used to select the most relevant genes in the dataset. In this way, this gene filtering reduces the number of genes to manage. Regarding the second step, the gene selection can be divided into two objectives: minimizing the number of selected genes and maximizing the classification accuracy. Therefore, we have used a multi-objective optimization approach. More exactly, an Artificial Bee Colony based on Dominance (ABCD) algorithm has been proposed for this second step. Our approach has been tested with eleven real cancer datasets and the results have been compared with several multi-objective methods proposed in the scientific literature. Our results show a high accuracy in the classification task with a small subset of genes. Also, to prove the relevance of our proposal, a biological analysis has been developed on the genes selected. The conclusions of this biological analysis are positive, because the selected genes are closely linked to the cancer dataset they belong to.