Benchmarking gene selection techniques for prediction of distinct carcinoma from gene expression data: A computational study

Abstract

Gene Expression (GE) data have been attracting researchers since ages by virtue of the essential genetic information they carry, that plays a pivotal role in both causing and curing terminal ailments. GE data are generated using DNA microarrays. These gene expression data are obtained in measurements of thousands of genes with relatively very few samples. The main challenge in analyzing microarray gene data is not only in finding differentially expressed genes, but also in applying computational methods to the increasing size of microarray gene expression data. This review will focus on gene selection approaches for simultaneous exploratory analysis of multiple cancer datasets. The authors provide a brief review of several gene selection algorithms and the principle behind selecting a suitable gene selection algorithm for extracting predictive genes for cancer prediction. The performance has been evaluated using 10-fold Average Split accuracy method. As microarray gene data is growing massively in volume, the computational methods need to be scalable to explore and process such massive datasets. Moreover, it consumes more time, labour and cost when this investigation is done in serial (sequential) manner. This motivated the authors to propose parallelized gene selection and classification approach for selecting optimal genes and categorizing the cancer subtypes. The authors also present the hurdles faced in adopting parallelized computational methods for microarray gene data while substantiating the need for parallel techniques by evaluating their performance with previously reported research in this sphere of study.