Analysis of polymorphic variation in drug metabolism: I. Kernal density estimation

Kernel density estimation was examined as an objective, nonparametric approach to the detection of polymorphic variation in distributions containing multiple complex data sets. Power curves were constructed for the kernel density estimate based on its ability to detect worked bimodality in stimulated distributions as a function of the distribution size, the fraction contained within a particular subdistribution, and the location of the mean of that subdistribution with respect to the mean of the overall distribution. Comparisons were then made between kernel density estimation and the Kolmogorov-Smirnov test of maximal differences. Results showed that kernel density estimation performed as well or better than the Kolmogorov-Smirnov test and offered a number of advantages, including identification of the frequency and placement of individual modes and antimodes. The Kolmogorov-Smirnov test, on the other hand, examined normality of a distribution rather than modality or inherent polymorphism, and the outcome was highly dependent on the subdistribution location and total distribution size. We conclude that kernel density estimation is an excellent method for analysis of polymorphic variation in drug metabolism.