What is the correct way of dealing with the analysis of data from flow cytometry?

Question

I would like to detect a change in expression of a molecule present on a cell type by flow cytometry.

Assuming I am able to detect, using an antibody, a signal that represents the amount of the molecule I'm interested into. Also assume that the same experiment will have some noise (ideally near to zero) represented by an isotype antibody. What if the noise is somehow higher than a minimal threshold (i.e. = 1%)? How can I deal with the noise? How can I compare different samples if the noise is higher than the threshold?

I was thinking about removing those samples that have a noise above the 1% but is this ok, from a statistical point of view?

Another possibility is: can I subtract the noise from the signal and use this information for the final statistical analysis? Is this correct? Is it possible to subtract a signal from another? How is this done?

EDIT just to clarify a bit:

I have a pool of neutrophils which, in resting condition, do not express the molecule on their surface. When stimulated, these cells are supposed to show an increased amount of this molecule (again, on their surface). We would like to quantify the change in the amount of molecule (we used FACS).

We are able to quantify the quantity of surface-molecule but we also get some aspecific signal (the noise due to the isotype antiboby). What we try to do is to keep the isotype signal lower than 1% using a gate-based methodology. What if the isotype signal we get is higher than the threshold (i.e. >1%)? The antibody sometimes work some others don't. What should we do when it returns too much noise? Should we discard the data or should we perform a normalisation over the surface-molecule using the noise?

Answer 1

If your goal is to simply "to quantify the change in the amount of molecule," then you can do that at the population-average level by simply taking the average antibody signal for the different cell populations (stimulated vs. not stimulated). In that case, you don't have to throw out samples with noise above a threshold - since they are still part of the population and so should be used when calculating average.

If you want to quantify "the change in the amount of molecule" at the single-cell level, then things are a bit more complicated, and the best answer depends in part on things like the shapes of the distribution of signal from the stimulated and non-stimulated populations. And related to that, when you say "noise above the 1%," do you mean that the Coefficient of Variation (CV) is greater than 1%? Or something else?