Analysis of polymorphic variation in drug metabolism: II. Effects of data transformation on the sensitivity and specificity of modal detection.

Simulations were conducted for a model of drug metabolism involving 2 parallel competing pathways of elimination, wherein the effects of variability in 1 enzyme pathway were examined with respect to variability in recovery of its metabolite and recovery of metabolite through the second, nonvaried, co-eliminating route. Expression of metabolite recoveries as fractions of the total recoverable drug yielded a statistic possessing variability similar to the pattern of variability induced in the enzyme itself. However, the transformation was subject to a "distributive" effect, in that the magnitude of variation in the downstream metabolite was reduced and transferred through reciprocal variations in availability of unmetabolized drug to other nonvaried pathways. The sensitivity and specificity of the fractional recovery statistic were thereby diminished. Expression of recoveries as metabolic ratios, on the other hand, limited variability to the pathway in which it was originally induced. The pattern of variation was skewed and exaggerated, particularly towards the rightward, "poor metabolizer" tail of the distribution, and this caused problems with visual interpretations as well as more objective approaches like the kernel density estimate. Additional transformation to the log metabolic ratio provided considerable improvement in this regard. Thus, log metabolic ratios are the most sensitive and specific of the data transformations and are the preferred manner of expression in all multipathway metabolite analyses.