Abstract
Biomarker discovery becomes the bottle-neck of personalized medicine and has gained increasing interest from various research fields recently. Nevertheless, producing robust and accurate signatures is a crucial problem in biomarker discovery and relies heavily on the used feature selection algorithms. Feature selection is a preprocessing step which plays a crucial role in omics data analysis to improve learning. The accumulating evidence suggests that ensemble methods and swarm intelligence are two growing solutions for improving feature selection algorithms. In this paper, we propose a two stages approach to identify a predefined number of biomarkers from gene expression data. It is designed as a wrapper-based ensemble method; each part of the ensemble is performed through cooperative parallel meta-heuristics and a filter-based mechanism. Experiments from twelve DNA microarray datasets have shown that our approach competes with and even outperforms recent state-of-the-art methods in terms of accuracy and robustness. Also, biological interpretation shows that our approach selects highly informative genes for cancer diagnosis.