FAQ  •  Register  •  Login

Discrimination between cell events-peak-finding algorithm?

<<

mleipold

Guru

Posts: 5796

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Wed Apr 23, 2014 4:05 pm

Discrimination between cell events-peak-finding algorithm?

HI all,

I have a general question about the cell-event peak-finding algorithm discrimination. The default settings to register a cell event are 3 standard deviations above background, with an event length of at least 10 pushes and no more than 75 pushes. This triggers when ions in at least one channel being measured satisfy these requirements.

However, if cell events are very close together so that the signal doesn't drop below the 3 S.D. threshold before 75 pushes before the second cell comes through, they may be read as a doublet. This is particularly an issue in the Ir channels, since all (or at least almost all) intact cells should register there.


But my question is this: if those two cell events are different types of cells (say, a T cell and a B cell), that differ greatly in their markers, there *won't* be overlap in a large number of channels....can the algorithm for "Found Cells" use those differences to split a T from a B, even if their Ir signals overlap?

I have attached a diagram to try to illustrate what I'm talking about. In that, I think all would agree that the software can separate Cell A from Cell B: the threshold between cell events is satisfied in all channels. My question is focused on Cell C from Cell D: the only channels that overlap are the DNA channels on the right.
Since the software is *monitoring* all the channels, I think this should be possible: in this case, a difference in any one channel would be enough to separate events (obviously, the more differences, the more confident the "call").

It's not clear to me from the last info I have on software versions. I have attached the release notes from "CyTOF Software V 5.1.648"; page 5 is the relevant page.


-Mike
Attachments
Peakfinding algorithm question.pdf
(15.96 KiB) Downloaded 437 times
CyTOF SWv648 release notes_2013-0326.pdf
(261.37 KiB) Downloaded 582 times
<<

robs

Contributor

Posts: 40

Joined: Mon Dec 02, 2013 3:42 pm

Location: University of Connecticut

Post Mon Apr 28, 2014 5:46 pm

Re: Discrimination between cell events-peak-finding algorith

I was hoping some people better versed in signal processing and .fcs generation would chime in first but here are a few thoughts, most of which you have probably already considered anyway Mike.

Flow events are usually treated as discrete events in time following Poisson statistics. Multiple events occurring at the same time (or very close) in flow come through as coincidences- the signal does not fall below threshold between the multiple events, happening because two or more cells are too close in the stream as they are being interrogated.

These coincidences are almost always thrown away either in initial signal processing as electronic conflicts or in the process of gating out doublets. So in regular flow these coincidences are pretty much treated as garbage, and it seems like the standard CyTOF algorithm treats them similarly by throwing away events with long event lengths. But maybe since the CyTOF actually follows a completely different triggering/thresholding algorithm maybe it can be tweaked to allow for single cell analysis of coincidences as you suggest.

A few issues that come to mind-

-If there is overlap in common channels such as DNA or CD45, how will that intensity signal be divided between the two events in the coincidence? proportionally? What if that common channel is an important signaling channel? Would we lose some accuracy on the expression on the individual cell? The parallel case in flow I can think of is trying to use modeling, like Mod Fit, to figure out the percentage of cells in each phase of cell cycle with only a DNA stain. The model can get us close but I always preferred to add other cell cycle dyes if possible to back up/inform my modeling.

How would we define this algorithm, since the signal processing now relies on breaking up the signals into discrete slices of time? This is something I have no idea about.

Does the frequency of these distinguishable coincidences justify an additional algorithm to process them? For instance, if there are only 1% coincidences (perhaps its many more) in a sample and only half of those are made up of two distinct cell types, are we gaining much more information with another algorithm?

Anyone else care to add??

-Rob
<<

antonio

Contributor

Posts: 20

Joined: Wed Dec 04, 2013 3:05 pm

Post Mon Apr 28, 2014 11:37 pm

Re: Discrimination between cell events-peak-finding algorith

Hi Mike,
the software does not make the separation between the event C and D, since while in the example you are showing it will be logical to consider two cells in a lot of other cases the splitting will be misleading.
Imagine a unique event/cell (30 push) that produces cytokines, you can have some positive pushes between push 2 and 10 for an IFNgamma signal and then some pushes positive let's say for the IL2 between push 20 and 25. In this case you do not want the alghoritm split your cell in two.
If you want to see the conversion system "live" you can transform the .imd in a text and play with it.
<<

anitamkant

Master

Posts: 51

Joined: Mon Nov 18, 2013 6:30 am

Post Tue Apr 29, 2014 2:34 am

Re: Discrimination between cell events-peak-finding algorith

Here is what happens in the case of C and D: Please see attached powerpoint for better explanation with figures.
A doublet event triggers a software subroutine designed to separate it onto two events. The events are too close to each other, and the second signal exceeds the threshold before the first one goes below the threshold. If a valley between events can be determined unambiguously, the doublet event will be split (assuming that the corresponding checkbox is enabled). Note that the leading tail of the second event will be included into the first event (integral) intensity.

When three cells are too close, it results in a triplet; and so on.
Attachments
Cell recognition question.pptx
(254.39 KiB) Downloaded 560 times
<<

mleipold

Guru

Posts: 5796

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Tue Apr 29, 2014 6:56 pm

Re: Discrimination between cell events-peak-finding algorith

Hi Anita and Rob,

Thanks for the reply, regarding the peak-finding in the case of overlap/doublets/triplets/etc. I see that the program definitely does not (currently?) do what I was hoping it would do.


My idea or question is that since the cell event can be triggered in any one channel, whether the program could use the fact that most channels *would* be resolved in satisfying the event criteria (3 S.D., event length) to increasingly separate doublets/multiplets.

In my example, C and D only have a valley/threshold problem in the far right (Ir-like) channels. But the other 4 marker-like channels with ions are completely resolved: C has no markers in common with D. In this case, using the 4 marker-like channels to "tell" the software that while the Ir channels *do* overlap (fail to resolve), that there *are* two cells there.


The issue of tailing would not be solved by this. However, in Anita's cytokine example, the two "new" cell events would fail to meet the 10 pushes threshold.


Rob's question about how to split up the ion signal is a good one. To a first approximation, you would give each "new" event the signal intensity in the pushes that define that event. So, as depicted at the top of page 2 of Anita's attachment, Event #1 would lose a bit of its right leg and gain a bit of #2's left leg. Event #2 would lose a bit of its left leg and gain a bit of #1's right leg. If these are ideal Gaussian peaks, though, those legs should cancel each other out. In a real-world setting, yes, I'd bet there would be a small amount of signal loss or gain for a given peak, relative to its "true" signal.

In a heterogeneous cell sample like PBMC, though, people generally run enough markers to resolve, say, a T cell from a B cell in a doublet. The only doublets that would get through would be two almost identical cells.
<<

jeccles

Participant

Posts: 5

Joined: Thu Jul 14, 2016 4:07 am

Post Fri Sep 21, 2018 3:19 am

Re: Discrimination between cell events-peak-finding algorith

Is it possible to ignore a given channel for cell event determination? My samples have a debris component that seems to be platelets based on the lack of DNA and presence of certain integrin markers. I would like to omit them from collection, which would seem possible if the few channels belonging to those integrin markers in my panel were not included in the event calculation. Please let me know if/how this can be done. Thanks!
<<

mleipold

Guru

Posts: 5796

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Fri Sep 21, 2018 4:32 pm

Re: Discrimination between cell events-peak-finding algorith

Hi Jacob,

Formally, yes, event acquisition can be changed from "all" channels to only some channels (or even just one).

In theory, you can edit this in the Custom Expression filtering option in your acquisition template (Helios manual screenshot attached).

However, to my knowledge, Fluidigm (or, originally, DVS Sciences) has never put out a guidance document on how to actually do this in practice. The closest I have seen was the original description in the CyTOFv1 manual: see attached CyTOFv1 manual screenshots. Even there, you may have to make a few attempts at reprocessing your IMD until you find the right threshold settings.


I've only ever heard of one person doing this, and that only for one experiment. I've heard that Fluidigm may have more info available if you contact your FAS, but nothing posted on their website.....


Mike
Attachments
CyTOFv2 manual-page 2.png
CyTOFv1 manual-page 1.png
Helios manual-custom expression filtering.png
<<

vtosevski

Contributor

Posts: 44

Joined: Wed Nov 20, 2013 12:50 pm

Location: Zurich, Switzerland

Post Mon Sep 24, 2018 11:48 am

Re: Discrimination between cell events-peak-finding algorith

Hi Jacob,

Yes, custom thresholding would be an option to consider. I have used it in the past to help go through an acquisition when one or more reagents were problematic. The caveat is that the noise reduction routine will work only in the channels defined in the custom expression. So, back then, I was thresholding on 140Ce and CD45 channel, getting beads and leukocytes thresholded on those two markers alone, with my other channels not having the benefit of the noise reduction routine being applied. Mike is right that this is not documented properly anywhere (to my knowledge). Back then I was getting this information from the DVS people directly.

Best,
Vinko
<<

mleipold

Guru

Posts: 5796

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Thu Nov 08, 2018 10:05 pm

Re: Discrimination between cell events-peak-finding algorith

Hi Jacob and Vinko,

While looking on Fluidigm's website for something else today, I found a document on the Custom Expression Filtering.

https://www.fluidigm.com/binaries/conte ... igm%3Afile

"Custom Channel Filtering with CyTOF Software v6.7"
"PN 400272 A1"


Mike
<<

vtosevski

Contributor

Posts: 44

Joined: Wed Nov 20, 2013 12:50 pm

Location: Zurich, Switzerland

Post Fri Nov 09, 2018 10:46 pm

Re: Discrimination between cell events-peak-finding algorith

Hi Mike,

thanks for sharing, it's interesting to learn more about the signal processing (didn't know preview and lagging offset are part of the imd file). The outlined procedure does not, however, describe the process we've been discussing in the earlier posts but, if I understood correctly, merely excludes the troublesome parameter from the processing and then applies default thresholding etc. A custom thresholding would keep all the channels (for good or for worse).

Vinko
Next

Return to CyTOF general discussion

Who is online

Users browsing this forum: AdsBot [Google] and 11 guests

cron