Computing the central location of immunofluorescence distributions: logarithmic data transformations are not always appropriate

Abstract

The idea of the "average" intensity of immunofluorescence data is often poorly defined, with such terms as average, mean, and peak used interchangeably. In addition, the common use of logarithmic amplifiers with immunofluorescence data further complicates the problem. Log amplifiers permit the display of a wider range of fluorescence intensities. At the same time, they effect a log transformation of the data. This transformation decreases the variance resulting in narrower fluorescence distributions, which are assumed to approximate normal distributions. When the log transform is used, the distribution mean is the geometric mean of the untransformed data, which is computed simply as the mean of the channel values. This mean value serves as a simple indicator of the population center. Despite the prevalence of log transformations in flow cytometry, this transformation may not yield normally distributed immunofluorescence data, whereas the square root or other fractional power transformations can yield normal distributions.