I have acquired data using FACS analysis and I'm curious how best to do statistics. For every donor (PBMC’s) I acquired 10 000 events for every separate culture setting and therefore I presume there to be a broad range of statistical possibilities. I have little experience with doing statistics on FACS data and therefore I wonder if anybody can give me some advice on the subject or recommend literature. (In the general immunological literature the information provided is very marginal)

Generally two parameters are used to describe individual samples: % positive cells and mean fluorescence intensity (MFI). Percent positive can be calculated in different ways. FlowJo gives several methods to calculate % positive events when overlapping histograms. Any of these methods, although having some mathematical validity, tend to give rather high numbers compared to the manual method of drawing a gate excluding most of the negative control events and asking how many events (less overlap of negative controls) are in your sample. MFI is a straightforward value calculated by the software. Other measures can be used, geometric mean, median etc, but MFI usually does the trick. Together these two measures give you a rough idea of how many positive events you have and how bright they are.

How you manage the statistical analyses of these individual descriptive statistics depends on the design of the experiment. The usual parametric and non-parametric analyses can be applied. If you have multiple samples of control versus test samples, then a simple unpaired t-test may suffice-applied to either the percent positive or MFI (which depends on what your trying to prove). It is hard to advise in too much detail on this matter without knowing the design of your experiment.

Thanks! But with the mean fluorescence intensity (MFI) you mean either the X mean or Y mean (depending of course what your looking at) provided in the programs statistics.

Which one of the MFI values you select will depend on what exactly you are looking for. For instance, if you are looking for expression levels of a particular surface marker, then the best thing to do would simply be to histogram plot the fluorescent one channel relevant to the marker and pull off the stats.

While the MFI values are fine, they should only really be used if your data is normally distributed. Most FACS data is present logarithmically and appears log normal (i.e it skews to either the left or right - normally the right). If your data looks "lognormal" then the geometric mean fluorescent intensity would give you a better indiciation of the centre than the arithmetic mean (the MFI). Once you have these numbers you can do some t-tests (I would guess in your case unpaired is probably what you'll need).

That said, I've seen some people use the median as an indicator of shift in FACS plots. The median is more resistant to outliers, as the median will not be skewed by them as much as the means.

This probably doesn't help to finish on this point though: which one you choose - either MFI, GMFI or median - will depend on what your data looks like.