A multi-objective optimization approach for the identification of cancer biomarkers from RNA-seq data

Identification of biomarkers is essential for the diagnosis and prognosis of certain diseases, like cancer. Gene selection purpose is finding the minimum number of genes that can classify a (e.g. normal or tumour) sample with a high accuracy. Therefore, the selected genes can be studied as potential cancer biomarkers. In this article, a new method for gene selection is proposed in two steps. The first step is a filtering of the most relevant genes of a gene expression dataset. In this step, three feature selection methods have been combined. Since gene selection is a two-objective problem (minimizing the number of selected genes while maximizing the classification accuracy), the second step is performed as a multi-objective optimization, using an Artificial Bee Colony based on Dominance (ABCD) algorithm. ABCD algorithm uses internally a support vector machine (SVM) classifier. The method has been tested with five RNA-seq cancer datasets and with a comparative study of the results obtained by the method and by other five methods proposed in the scientific literature by other authors. Finally, in order to check if the genes selected by the proposed method could be studied as biomarkers, the relation between the selected genes and the cancer they belong to is analysed. It can be concluded that the proposed method is effective in gene selection for the identification of cancer biomarkers from RNA-seq data.