A Metaheuristic Approach for Simultaneous Gene Selection and Clustering of Microarray Data

Cancer subtype discovery from gene expression data is a daunting task due to the relatively large number of genes associated with the samples. Selecting important genes to identify the underlying phenotype structure has been addressed in this work. The proposed work integrates gene selection using Firefly algorithm into K-means clustering. The algorithm guarantees a suboptimal subset of genes necessary for disease subtype clustering. Experiments were conducted on publicly available cancer gene expression datasets and an accuracy of 70-80% was obtained. Comparison with the previous work on Particle Swarm Optimization based feature selection shows that the proposed work achieved better convergence and accuracy.