Hi Katie,
In your post, you say that historically you've used Pop X Count to set a denominator threshold for a Freq calculation (eg, you're OK with reporting a number for PopX/Parent but not Child/PopX when PopX Count is low). You've historically set the threshold at Count=100, and you're wondering if you can set it lower (Count=20-40).
I think this is partly related to a recent thread on cell acquisition/number here:
viewtopic.php?f=1&t=2670In it, Cynthia Guidos does a good job discussing things like CV thresholds and therefore cell number needed. In particular, "Lower precision (10-20%) is generally acceptable for rare event analysis when it is not feasible to collect enough events to achieve higher precision."
There's also this article ( "Technical issues: flow cytometry and rare event analysis" PMID: 23590661 ) which discusses this in Table 1. I think you need to talk with your statisticians and decide on the lowest threshold (CV) that you're going to allow. Looking into statistical decisions made on RNAseq data might be a good thing too: in almost every case (cost, etc), they are dealing with far fewer cells than you would ever have in a cytometry experiment.
I guess there's a second question with regard to your MMI question: are you also looking for a separate confidence threshold for the Metal Intensity value you report, or are you simply asking what Freq CV threshold you'd need in order to report *any* MMI value? With Freq CV, you're talking about the confidence of robustly detecting and properly quantifying the *presence* of something. With Metal CV, you're talking about the confidence of assigning a signal intensity value. If you think about this in the realm of a highly uniform sample like beads, it would be the difference between spiking the beads in at different concentrations (Freq CV), versus the signal intensity distribution of the metal signal in the detected beads (Metal CV).
Formally, I'm not sure that the Freq CV would scale at exactly the same rate as the Metal CV: in the bead example, the signal intensity on the beads is highly uniform (tighter distribution), so I would think that you would need fewer events for a given SD/CV threshold for the Metal Intensity. In the case of a cell sample, though, I would assume that biological variability (broader distribution) would lead to a greater Metal Intensity variability. This would depend on the marker in question: for example, CD27 is fairly tightly bimodal on T cells with a small MMI distribution in both CD27pos and CD27neg populations, whereas CD38 expression is much more smeary (broader distribution).
Mike