Let me add one comment to this... Maciej is right that minor miscompensation will not affect your results if you are gating on easily-separable populations. It can strongly affect your results if you are gating on difficult-to-separate populations ("smears"). Most importantly, it will INVALIDATE your results if you are relying on MFI (fluorescence intensity) calculations. For those who are doing MFI analyses, it is absolutely essential to validate your compensation settings! mr (PS, it is really encouraging that I saw so many excellent explanations of compensation on the mailing list recently! Really outstanding.) On Feb 11, 2008, at 11:54 AM, Maciej Simm wrote: > Hello Li, > > I want to emphasize that looking at this is by no means the > definitive > benchmark of proper compensation, but can be used to spot problems. > As > others pointed out beads will help you do better comps. My point was > simply this - if your events are symmetrically distributed in a > funnel > shape created by the exponential distribution, your compensation is > probably done correctly. If you notice that events are biased away > from this symmetry either inward (undercompensated) or outward > (overcompensated) you should scrutinize your compensation tubes to > see > what caused this bias. > > In many mutually exclusive (CD3 VS CD19) or strongly bimodal reagent > combinations (CD3/4/8) you can get away with small compensation > problems - simply because the clusters are so far away, slight > errors will not affect your ability to properly resolve them. In the > context of these distributions, the answer to your question is "it's > probably not going to affect your final analysis very much". > > However, if you are looking at more difficult-to-resolve > distributions > like cytokines or activation markers smearing across the entire > dynamic range - it becomes VERY important that the comps are right, > because a slight comp error could translate into a much bigger > analysis error than in the previous scenario, because it may not be > clear where the line should be drawn to separate "positives" from > "negatives". > > Maciej > > > > > On Feb 10, 2008, at 7:36 AM, chen li wrote: > >> Hi all, >> >> >> Just a follow-up question about Maciej' graphs. >> >> If I get some graphs showing either a little bit over >> or under compensation similiar to those in Maciej's >> graphs, how these are going to affect my final >> analysis, compared to the perfect compensation one? >> >> Thanks, >> >> Li >> >> >> --- Maciej Simm <simm@treestar.com> wrote: >> >>> Colleagues, >>> >>> It's very difficult to answer the question "should I >>> do it by hand or >>> computer" without the whole context of the specific >>> scenario when one >>> fails and the other works. >>> >>> Instead, we can focus on a broader question "is my >>> compensationcorrect?" >>> >>> Mathematically, we can estimate validity of our >>> compensation matrices >>> by comparing medians - for example, the median of >>> the positives is >>> equal to the median of the negatives. Medians are >>> better than mean/ >>> GM's because they aren't affected by outliers and >>> better represent >>> central tendency of your data. >>> >>> Since biexponential transformation (aka logicle >>> display, etc..) became >>> popular in many software packages, there is an >>> easier way to 'eyeball' >>> bad compensation - enable this scaling and see if >>> your single- >>> negatives (some call them single-positives, I'm a >>> sheath tank half- >>> empty kinda guy ;) ) are symmetrically distributed >>> around the X=0 / >>> Y=0 lines similar to your double-negatives. If not, >>> there's either >>> over comp, or undercomp. Here's an example, with >>> hypothetical X/Y >>> parameter distributions: >>> >>> http://cd4cd8.com/1.png proper comp -1 % - the X+/Y- >>> population leans >>> upward, suggesting undercompensation. >>> http://cd4cd8.com/2.png proper comp +1% - the X+/Y- >>> population is >>> sagging toward X axis. Over comp'd. >>> http://cd4cd8.com/3.png proper comp (correct values) >>> - the centers of >>> X-/Y- and X+/Y- are aligned. >>> >>> hope this helps, >>> >>> >>> Maciej Simm >>> TreeStar Inc. >>> >>> >>> >>> >>> >>> >>> PS. I vote for pistachio ice cream. >>> >>> >>> >>> >>>>> I have people in the lab who complain about me >>> not doing manual >>>>> compensations and instead have the Aria calculate >>> them for me. And >>>>> when I have the Aria compensate, they want to >>> manually compensate >>>>> after they record the data because the plots >>> "don't look right." I >>>>> mean, isn't that simply producing "make-belief" >>> data? Or am I >>>>> missing something? >>> >> >> >> >> >> ____________________________________________________________________________________ >> Be a better friend, newshound, and >> know-it-all with Yahoo! Mobile. Try it now. >> http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ >> >>Received on Wed Feb 13 15:18:00 2008
This archive was generated by hypermail 2.1.8 : Wed Jan 31 2007 - 03:12:00 EST
