Minimizing Redundancy Among Genes Selected Based on the Overlapping Analysis
For many functional genomic experiments, identifying the most characterizing genes is a main challenge. Both the prediction accuracy and interpretability of a classifier could be enhanced by performing the classification based only on a set of discriminative genes. Analyzing overlapping between gene expression of different classes is an effective criterion for identifying relevant genes. However, genes selected according to maximizing a relevance score could have rich redundancy. We propose a scheme for minimizing selection redundancy, in which the Proportional Overlapping Score (POS) technique is extended by using a recursive approach to assign a set of complementary discriminative genes. The proposed scheme exploits the gene masks defined by POS to identify more integrated genes in terms of their classification patterns. The approach is validated by comparing its classification performance with other feature selection methods, Wilcoxon Rank Sum, mRMR, MaskedPainter and POS, for several benchmark gene expression datasets using three different classifiers: Random Forest; k Nearest Neighbour; Support Vector Machine. The experimental results of classification error rates show that our proposal achieves a better performance.
- Apiletti, D., Baralis, E., Bruno, G., & Fiori, A. (2012). Maskedpainter: Feature selection for microarray data analysis. Intelligent Data Analysis, 16(4),717–737.Google Scholar
- Gordon, G., Jensen, R., Hsiao, L., Gullans, S., Blumenstock, E., Ramaswamy, S., et al. (2002). Translation of microarray data into clinically relevant cancer diagnostic tests using gene expression ratios in lung cancer and mesothelioma. Cancer Research, 62(17), 4963–4967.Google Scholar
- Kestler, H., Lindner, W., & Müller, A. (2006). Learning and feature selection using the set covering machine with data-dependent rays on gene expression profiles. In F. Schwenker & S. Marinai (Eds.), Artificial neural networks in pattern recognition (ANNPR 06) volume LNAI 4087 (pp 286–297). Heidelberg: Springer.Google Scholar
- Mahmoud, O., Harrison, A., Perperoglou, A., Gul, A., Khan, Z., & Lausen, B. (2014b). propOverlap: Feature (gene) selection based on the proportional overlapping scores. R package version 1.0, http://CRAN.R-project.org/package=propOverlap
- Mahmoud, O., Harrison, A., Perperoglou, A., Gul, A., Khan, Z., Metodiev, M., et al. (2014a). A feature selection method for classification within functional genomics experiments based on the proportional overlapping score. BMC Bioinformatics, 15, 274.Google Scholar