Friday, April 17, 2009

What is MFI?

If you've read any papers with flow cytometry data in it, undoubtedly you've come across the abbreviation, MFI. Generically, people expand this to Mean Fluorescence Intensity, but ironically, you'd rarely use the actual Mean of the population. Basically what the MFI is suppose to measure is the shift in fluorescence intensity of a population of cells. In cases where the entire population stains with different levels of an antibody (like measuring expression level of antigen x), it would be appropriate to report relative MFI values based on some sort of control (unstained, isotype, FMO, etc...) to demonstrate an increase or decrease in expression of this marker (assuming that each sample was stained with saturating amounts of antibody, and all samples were run under the same conditions and instrument settings blah, blah, blah). So, if you wanted to make measurements like this, what statistics would you use? When you analyze your data in software (e.g. FlowJo) you are given options to calculate the Mean, Median, Mode, and Geometric Mean. I've included a link which explains these measures in terms of flow cytometry data pretty well, so i won't bother going through that here. But, I will give you the punchline. When in doubt, use Median Fluorescence Intensity. Mean is pretty much useless, it doesn't work too well on a log scale, and for non-normal distributions, it is easily affected by outliers. I don't mean to be so mean when talking about the mean, but hey, for flow data on a log scale, why bother (sorry, i couldn't resist with the 'mean' pun). If you feel you must use an arithmetic average on a log scale, use Geometric Mean.