Hybrid SPR algorithm to select predictive genes for effectual cancer classification

Authors: ARUNA SUNDARAM, NANDAKISHORE LELLAPALLI VENKATA, RAJAGOPALAN SARUKAI PARTHASARATHY

Abstract: Designing an automated system for classifying DNA microarray data is an extremely challenging problem because of its high dimension and low amount of sample data. In this paper, a hybrid statistical pattern recognition algorithm is proposed to reduce the dimensionality and select the predictive genes for the classification of cancer. Colon cancer gene expression profiles having 62 samples of 2000 genes were used for the experiment. A gene subset of 6 highly informative genes was selected by the algorithm, which provided a classification accuracy of 93.5%.

Keywords: Cancer classification, filters, wrappers, correlation feature selection, sequential backward search, support vector machines, DNA microarray

Full Text: PDF