Journals Proceedings

Abstract

Attempts are still in full swing for the detection of differentially expressed genes and using the genes for the detection of early cancer. Each method is with its own simplicity or complexity with the results, yet to portray some degree of confidence level in the designation of a gene as being differentially expressed or not expressed. However, no one statistic is universally optimal and there is seldom any basis or guidance that can direct toward a particular statistic of choice for such complicated gene data. Identifying a subset of genes that are expressed differentially in leukemia types from a large pool of candidates’ genes generated by Golub et al(1999) in microarray experiment has been empirically demonstrated in the study. Principal component analysis on a Box-Cox transformed data exposed clusters of specific leukemia types. Taking into consideration the consistency of a gene expression the relative variance of each gene across the sample expression profiles is determined. A minimum threshold value of gene expression level from the data set developed a feature selection discriminant rule that discriminated genes into their specific leukemia types. Quadratic discriminant analysis provided encouraging results, validating identification of genes with probability of correct identification exceeding 85%. The group of differentially expressed genes, when subjected to principal component clustering fell into clusters of their own specific leukemia types.