An Experimental Study on Microarray Expression Data from Plants under Salt Stress by using Clustering Methods

Current Genome-wide advancements in Gene chips technology provide in the “Omics (genomics, proteomics and transcriptomics) research”, an opportunity to analyze the expression levels of thousand of genes across multiple experiments. In this regard, many machine learning approaches were proposed to deal with this deluge of information. Clustering methods are one of these approaches. Their process consists of grouping data (gene profiles) into homogeneous clusters using distance measurements. Various clustering techniques are applied, but there is no consensus for the best one. In this context, a comparison of seven clustering algorithms was performed and tested against the gene expression datasets of three model plants under salt stress. These techniques are evaluated by internal and relative validity measures. It appears that the AGNES algorithm is the best one for internal validity measures for the three plant datasets. Also, K-Means profiles a trend for relative validity measures for these datasets