RE: gate-specific data cropping

Stevan

At the risk of sounding like a commercial, WinList can do exactly what you
want.  The problem of "sucking up air" or perhaps having a clogged nozzle at
the end of a run has gotten all of us at one time or another, and then the
data (most of which is probably fine) appears to be unusable.  WinList
addresses this issue with the FTIM parameter, which is basically a
"chronology" parameter defined over 1024 channels.  Each event is assigned
an FTIM channel based on its order in the listmode file, channel zero having
the first n/1024 events, and channel 1023 having the last n/1024 events.

Displaying a 2P dot plot of FTIM vs. side scatter, for example, will clearly
show the aberrant events.  You may then gate on the "good" events, or gate
out the "bad" events.  Either way, you can salvage an otherwise bad run. 

You can also save the gated listmode using WinList's Save Data Source
option, and it will contain only
the "good" events.  Another feature of the Save Data Source option is the
ability to set the number of events to save in the listmode file, which at
least partially addresses the desire to save a larger file into smaller
segments that will each have the same number of events within given regions,
as you are requesting.

Best regards,
Don
Donald J. Herbert
Technical Support Manager
Verity Software House


-----Original Message-----
From: Stevan Lauriault [mailto:stevan@lauriault.com]
Sent: Friday, February 09, 2007 6:24 PM
To: cyto-inbox
Subject: Re: gate-specific data cropping

Thanks for your very helpful comments.	I guess what I am trying to get at
is this: is
there any analysis software available that can reduce list mode data in a
reverse order-specific matter (in reverse sequence)?  Correct me if I'm
wrong, but I believe this function would also help in the event that a user
lets a sample run dry and sucks up air bubbles, or over collects their
intended gating target (for example, in acquisition and storage, 5000 of
R1).

Thanks,

Stevan
---------- Original Message ----------------------------------
From: "Byron Ellis" <bcellis@stanford.edu>
Date:  Wed, 7 Feb 2007 16:21:28 -0800

>On 2/7/07, Stevan Lauriault <stevan@lauriault.com> wrote:
>> Hi and thanks,
>>
>> Since we want to directly compare MFIs (in FL2) between any two 
>> subsets (R1 vs. R3)
>that are isolated on a bivariate dot plot (FL1 vs. FL4), doing so in 
>separate sample tubes would introduce the extra probability of 
>experimental
error between sample tubes.
>Directly comparing the subsets from the same sample tube, and even 
>better, the same data set, would eliminate this extra unknown.
>
>It doesn't necessarily eliminate the unknown, it may just keep you from 
>finding out about it. If you were to prepare one sample on Monday, see 
>significant differences, celebrate, etc and then on Wednesday prepare 
>another sample where you didn't you'd probably want to know that (and 
>figure out why). Simply taking the FACS tube off the cytometer and 
>putting it back on won't give you that information. Now, would I be 
>surprised if that actually happened in your case? Yes.
>Would I do replicates anyway? Yes.
>
>>
>> Let us say we have acquired 100,000 total events (stained with 3 
>> fluorochromes,
>measured in FL1, FL2, and FL4 respectively).  We are using a bivariate 
>dot plot of FL1 vs. FL4 in which we are isolating 3 subsets (R1, R2, 
>and R3) based on their relative expressions of FL1 vs. FL4.  We then 
>want to compare the MFIs, as measured in FL2, among these subsets.
>Within those 100,000 total events, there are 500 of R1, 700 of R2 and 
>1200
of R3.
>>
>> If we crop the 100,000 total events to 75,000 total events (from the 
>> top in
>stack-collected data), there will become exactly 500 events in the R2
region. If we
crop
>the 100,000 total events to 40,000 total events, there will become 500
events in R3.
>>
>> Now the three populations; R1, R2, and R3 each have exactly 500 total 
>> events, and we
>can directly compare MFIs among the subsets that have identical sample
sizes.	We could
>also then say that, when comparing means between these subsets under 
>identical experimental conditions, Rx gives a consistently higher
fluorescent signal than Ry.
>
>Well, depending on what you're doing (you've never said how you intend 
>to compare means), equal sample sizes is not particularly required so 
>there's no particular reason to cut the data to obtain equal sample 
>sizes. For example, one-way ANOVA (or one-way ANOVA with repeated 
>measure in the event that you chose to do _real_ replicates), which 
>simply says "they're different" (there are other tests, variants of 
>one-sided t-tests essentially, that give you directionality statements 
>such as mu_{R_1} > mu_{R_2} ). In this case your assumptions are:
>
>1. Each group is independent
>2. Normality of group populations (or at least "close enough" to
>Normal) 3. The _population_ variances of the dependent variable for 
>each group are equal (or pretty close). (Actually check this. It would 
>be unsurprising to encounter large differences in the population 
>variances among your different groups).
>
>Like I say, it depends a bit on what you intend to do for your 
>analysis. For example, _two_-way ANOVA actually expects equal sample 
>sizes.
>
>>
>> Is this a reasonable strategy?  Is there any software available that 
>> can perform this
>function?
>
>On an FCS file directly? Probably not (neither the manipulation of the 
>flow data nor the testing). Once you've got the data out of FCS form 
>and appropriately labeled with group membership or split into separate 
>files or something similar, then any reasonable statistics package 
>(Excel is not included in the list of reasonable statistics packages) 
>can perform the statistical tests.
>
>Personally, I use R (http://www.r-project.org) for analyzing all the 
>flow data I have, but it can be intimidating for new users (though 
>extremely powerful once you learn to use it). There are two packages in 
>R for manipulating flow data at the moment (prada and rflowcyt), 
>available through the Bioconductor Project
>(http://www.bioconductor.org) that can actually read in and manipulate 
>FCS data directly or slightly manipulated data from say FlowJo (so you 
>can do your gating there for example). I'm also part of a group made up 
>of the people who did rflowcyt and prada that are making a combined 
>package of the best bits of both (and some other new bits) that we 
>expect to be available in the next Bioconductor release (probably 
>sometime in April).
>
>Hope that helps,
>
>Byron
>
>>
>> Kind Regards,
>>
>> Stevan Lauriault
>>
>>
>> ---------- Original Message ----------------------------------
>> From: "Byron Ellis" <bcellis@stanford.edu>
>> Date:  Tue, 6 Feb 2007 12:17:35 -0800
>>
>> >Speaking as a statistician, I would not count running the same tube
>> >3 times or simply cutting the FCS file (the two are equivalent) as 
>> >three replicates so I wouldn't do either to achieve what you want 
>> >(though I can imagine situations where you would resample to 
>> >calculate statistics). To get three replicates you would have to 
>> >prepare three separate samples---you're interested in the 
>> >variability of the mean due to experimental error as well as the 
>> >biological variability of the cells and a single sample preparation 
>> >only
captures the latter.
>> >
>> >
>> >
>> >On 2/5/07, Stevan Lauriault <stevan@lauriault.com> wrote:
>> >> Dear All,
>> >>
>> >> Question:
>> >>
>> >> We are isolating leukocyte subsets to measure and compare relative 
>> >> levels of
>expression
>> >> of certain antigens.  For example, there are three subsets within 
>> >> a bivariate dot
>plot;
>> >> gated on R1, R2 and R3 respectively, and we would like to 
>> >> statistically analyze the relative mean fluorescence intensities 
>> >> of a
biomarker, among the subsets.
>> >>
>> >> To get statistically comparable sample sizes among these subsets, 
>> >> we are running
the
>same
>> >> tube three times and changing the storage criteria to 500 events 
>> >> for each subset (example, the first run is 500 events of R1, the 
>> >> second 500 of R2, and the third is
>500
>> >> of R3).  Instead of repeating the same tube three times, and 
>> >> wasting valuable
>sample, it
>> >> seems like a good idea to "crop" a preexisting data file of, for 
>> >> example, 100,000
>total
>> >> events, three times to get an identical sample size for all three 
>> >> gated subsets
(R1,
>R2,
>> >> and R3).
>> >>
>> >> Scientifically, this shouldn't be a problem, since flow cytometry 
>> >> data is collected randomly within the same sample tube.  Not only 
>> >> could we run one acquisition to get different monocyte subset 
>> >> populations, but also to statistically compare constant
>numbers
>> >> of monocytes, PMNs, lymphocytes, lymphocyte subsets, etc. However, 
>> >> I can see why
>some
>> >> scientists might be uneasy with the idea of manipulating raw 
>> >> list-mode
data.
>> >>
>> >> However, Since the data is collected randomly from the same 
>> >> sample, doing this
>should
>> >> give exactly the same readings as if we just collected less 
>> >> events, since we would
>only
>> >> be cropping (from stack-collected data) the most recent events
collected.
>> >>
>> >> Currently, we are acquiring 3 different data files for each tube, 
>> >> and adjusting the acquisition and storage settings for each 
>> >> acquisition.  Is there a way to perform a gate-specific data 
>> >> cropping function with any current flow cytometry data analysis
software?
>> >>
>> >> Kind Regards,
>> >>
>> >> Stevan Lauriault
>> >>
>> >>
>> >>
>> >>
>> >> ________________________________________________________________
>> >> Sent via the WebMail system at lauriault.com
>> >>
>> >>
>> >>
>> >
>> >
>> >--
>> >Byron Ellis (byron.ellis@gmail.com)
>> >"Oook" -- The Librarian
>> >
>>
>>
>>
>>
>>
>> ________________________________________________________________
>> Sent via the WebMail system at lauriault.com
>>
>>
>>
>>
>>
>
>
>--
>Byron Ellis (byron.ellis@gmail.com)
>"Oook" -- The Librarian
>




________________________________________________________________
Sent via the WebMail system at lauriault.com