Dear all,
I need your feedback on the right way to analyze flow data for my experiment. I have read articles in a number of blogs and discussion forum about the usage of mean vs median fluorescent intensity in FACs data and most people have voted for Median FI to measure relative fluorescence. I also agree with them. However, I believe choosing mean or median to analyze flow data is context dependent.
In my experiment, I have a bimodal GFP distribution with two distinct peaks- one at a low fluorescent intensity range while the other at a high fluorescent intensity range. I am monitoring the drop in GFP over time.
In my case, the experiment starts with only the high-GFP peak and gradually with time, the Low-GFP peak emerges and the high-GFP peak falls. But there is no shift in the peaks. Eventually, the high-GFP peak disappears and the low-GFP peak dominates.
In this context, the median or mid value stays the same, but the mean changes. I think, in this case, the arithmetic mean is the right choice for estimating GFP down-regulation over time.
Any expert opinion?
Please refer to the attached ppt file. I have drawn a cartoon diagram to explain the problem.
Hi Sutanuka,
I agree with you that the mean and the median have their benefits. The median is less sensitive to outliers, therefore it won't be shifted by some outlier cells whose intensity is distorted by noise. However, if these outlier cells are those you are interested in, then the median will obscure the effect.
Regarding your specific problem I think the median must change when the relative contribution of the two populations changes in time, because the median is NOT the value halfway between the lowest and highest intensity. Assuming you have 10000 cells in the histogram, and start counting the cells FROM THE LOWEST INTENSITY, the median is the intensity of the 5000th cell.
In order to characterize the increase in the low-GFP subpopulation in the best way, I would fit two Gaussian peaks on the histogram, and the fit would give you the relative fraction of cells in the low- and high-GFP peaks. You can do it in many different programs (Matlab, Origin, but even Excel will do it for you after a little bit of tweaking). Alternatively, if the two peaks don't overlap significantly, then you simply place gates or markers on both of the peaks, and determine the number of cells in the two subpopulations, and report the relative fraction of cells in one of them. If the two peaks significantly overlap, this latter approach could lead to significant miscalculations.
Good luck,
Peter
Considering the nature of the data, the median and mean are not sensible descriptors. You should make a graph of the proportion of cells in the high-fluorescence and low-fluorescence peaks as a function of time. Since the peaks are well-separated, it should not be a problem to calculate these proportions.
Thanks, Peter and Adam. Your suggestions were immensely helpful
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Physical Phenomena
- Luminescence
- Fluorescence