Responses to " automatic compensation"

Rossi Ralph ralph.rossi at
Fri Jun 29 03:18:34 EDT 2007

Thanks to the many respondents , below is an edited list of replies.

Ralph Rossi
Flow Facility Manager
Peter MacCallum Cancer Centre
r.rossi at
Ph 96561955


1) I trust autocomp with beads during QC, I trust it for basic dye set-ups - FITC, PE, PerCP, APC.
For more than 4 colours, for conjugated dyes (APC-Cy7 etc), I like to check. I have one research group who do all their comps (usually 5-7 colour) manually.

2)we recommend checking automatic spillover results, at least against what "looks" right. In a Polycromatic flow experiment (i.e., >4 colors), It's always a good idea to look at cross-color interactions, so by default you will re-evaluate spillover. 
In most cases, spillover calculations will be accurate. Where it becomes challenging is when using a) tandems, and b) low "brightness" fluors. Complex spillover interactions will require occasional "feathering" of values . . . a subjective practice by default. Careful selection / pairing of colors, minimizing tandems (using the new -H7 are an improvement), and good instrument characterization (i.e., setting PMT voltages correctly) will also help. 

3)The basic algorithm of matrix inversion to calculate compensation has been known since the first published paper on compensation by Loken, Parks, et al.
However, specific implementations might vary, for example,, adding noise to the baseline, etc.
So I think it is always a good idea to check the results on your compensation controls.

4) Auto compensation system could be trusted. But it really depend of what you use for compensation. You need to use single color tube for each fluorescence parameter you have. Depending of what you look, the choice of these single color tube could be really important. Basically, you need the same material for auto-compensation that you need for manual compensation. If you try to post compensated data, without having single color files with the specific voltages and gain using during your experiment, it won't work. 
So no matter what do you do for comp, you need to have the proper control.

5) I always teach customers how to validate the autocomp setup as it is the same as teaching manual . The difficulty is not as much in the algorithm as in the gate setting on the control files. 
6) YES, if you run proper controls. Single colors samples of EACH fluor, (that means each individual tandem if you use them too!) with negative cells in the tubes as well.... BUT you can easily check if they are calculating correctly by drawing a box around the negative population and the positive in the bivariate plot and seeing if the medians are the same. Simple check.
7)In theory, the automated process is less prone to judgment artifacts. In practice, the problem is that the integrity of the autogenerated matrix is wholly dependent on the quality of the single-color controls that you run. If there are problems with the controls (e.g. inadvertent double staining, contamination by other cells, a heterogeneous control sample with fluorescent emission shifts in one sub-population versus another, etc.), it'll throw the matrix off. I'd argue that it should be almost mandatory that the user inspect and, if necessary, modify the populations that are being employed to to calculate the matrix to make sure there are enough cells, they are the cells they think they are, that there are no wild outliers being included, etc. Without a "manual override" provision, I'd be very suspect of any algorithm. 

-------------- next part --------------
HTML attachment scrubbed and removed

More information about the Cytometry mailing list