Robustness of the Chen-Dougherty-Bittner procedure against non-normality and heterogeneity in the coefficient of variation.

Abstract

Chen, Dougherty, and Bittner [Y. Chen, E. R. Dougherty, and M. L. Bittner, J. Biomed. Opt. 2(4), 364-374 (1997)] provided the derivation of a probability density function (PDF) for a signal ratio from a DNA microarray. This PDF is potentially useful for testing whether a pair of signals from the same gene has a common mean. The derivation of the PDF assumes the normality of all signal distributions and a common coefficient of variation (CV) for all signals within a microarray. The testing procedure requires the calculation of a common confidence interval for a microarray, based on a maximum likelihood estimator of the "common" CV, and the determination of whether or not a ratio for a particular gene falls within this interval. This study used Monte Carlo techniques and demonstrated that the procedure is robust to violations of normality and also to constancy in the coefficients of variation. A closer examination of the dynamics of the procedure found that the robustness was the result of offsetting effects. The size of the confidence interval was increased as a result of higher estimates of the common CV, as the actual CV pattern became heterogeneous. This effect mitigated the inflation in the size of the ratio as a result of increasing CV heterogeneity. These findings suggest that the Chen-Dougherty-Bittner procedure may be used even if underlying assumptions do not hold.