FAQ  •  Register  •  Login

Verity Cen-se?

Forum rules
Please be as geeky as possible. Reference, reference, reference.
Also, please note that this is a mixed bag of math-gurus and mathematically challenged, so choose your words wisely :-)
<<

kunicki

Contributor

Posts: 38

Joined: Thu Apr 13, 2017 8:46 pm

Post Wed Oct 25, 2017 11:34 pm

Re: Verity Cen-se?

I would think the aim should be to show the technique (1) has less bias than other singlet gating methods, and (2) provides more consistency between replicates and experiments. Both are equally important.

In theory, it would be very tough to prove that this method captures close to all phenotypic data possible, with little to no bias, because one might ask what would make our best biological standard? Does this standard cover everything? Still, the evidence Bagwell shows regarding the difference in DNA-Ir signal between different cell populations gives us a real benchmark to compare "manual" (and dare I say?) "pulse" gating techniques. Say we deplete a sample of myeloid cells, and spike in a known amount, to see how accurate the two gating methods are (...for example)?

Without a doubt, this would be a *fun* publication to write. :D

Best,
Matthew
<<

dtelad11

Master

Posts: 129

Joined: Mon Oct 31, 2016 6:26 pm

Post Wed Oct 25, 2017 11:40 pm

Re: Verity Cen-se?

kunicki wrote:Still, the evidence Bagwell shows regarding the difference in DNA-Ir signal between different cell populations gives us a real benchmark to compare "manual" (and dare I say?) "pulse" gating techniques.


I might have missed something or misunderstood you, which evidence is that?

kunicki wrote:Without a doubt, this would be a *fun* publication to write. :D


I am guessing that you are being sarcastic :-) As someone whose specialty is method development, I am completely fine with writing method papers. Our recent AOF paper (http://www.sciencedirect.com/science/ar ... 5917303836) doesn't have any exciting biological insight, but I feel that it is a helpful resource (at least based on the follow-up emails we already received). And honestly it was pretty fun to write! We have two other papers in the pipeline.

As an aside, the seminal CyTOF papers are method papers ;-)
<<

AdeebR

Grand master

Posts: 169

Joined: Thu Mar 13, 2014 5:58 pm

Location: NYC

Post Wed Oct 25, 2017 11:57 pm

Re: Verity Cen-se?

Just an FYI that I've got some benchmark datasets in the works (which I was planning to make public in the near future) that I've designed for the exact purpose of conducting these kinds of tests of different doublet exclusion strategies.

Adeeb
Adeeb Rahman
Icahn School of Medicine at Mount Sinai, NYC
<<

kunicki

Contributor

Posts: 38

Joined: Thu Apr 13, 2017 8:46 pm

Post Thu Oct 26, 2017 12:00 am

Re: Verity Cen-se?

Hi El-ad,

In his presentation, he showed a Cen-se' diagram of two large populations, both of which did not share characteristics similar to bead, debris, dead cell, or "doublet" events. These two populations differed in mean DNA-Ir signal, similar to what you might see if you back-gate onto DNA-Ir(191/193) from Myeloid (CD3- CD19- HLA-DR+) or Lymphoid (CD3+ or CD19+) populations. I think this difference in DNA-Ir signal between populations came up earlier in the forum in a separate thread. Unfortunately, Bagwell is not defending his argument with full scientific rigor, which I think is where all our anxiety comes from. In theory, the technique 'makes sense', but is this technique 'truly' better at identifying singlets? I don't know yet. If one technique consistently cuts out higher DNA-Ir signal singlet populations, and the other doesn't, then I think we can at least use this as one benchmark.

Barcoding could also help us better assess if these events are singlets, as Greg mentioned earlier. There is a lot to consider. No sarcasm here. I really imagine it would be fun to write. :D

Matthew
<<

dtelad11

Master

Posts: 129

Joined: Mon Oct 31, 2016 6:26 pm

Post Thu Oct 26, 2017 12:06 am

Re: Verity Cen-se?

kunicki wrote:In his presentation, he showed a Cen-se' diagram of two large populations, both of which did not share characteristics similar to bead, debris, dead cell, or "doublet" events. These two populations differed in mean DNA-Ir signal, similar to what you might see if you back-gate onto DNA-Ir(191/193) from Myeloid (CD3- CD19- HLA-DR+) or Lymphoid (CD3+ or CD19+) populations.


Okay, that's what I thought you meant! To be blunt, I would not call that "evidence" ... or anything that relies on a dimensionality reduction method followed by eyeballing ...

kunicki wrote:If one technique consistently cuts out higher DNA-Ir signal singlet populations, and the other doesn't, then I think we can at least use this as one benchmark.


But how can we use this as a benchmark? As far as I know there is no quantitative evidence that high DNA-Ir has anything to do with doublets! We're just following Bendall et al's 2013 work.

kunicki wrote:No sarcasm here. I really imagine it would be fun to write. :D


Then it appears that enjoying writing method papers is something we have in common :-)
<<

kunicki

Contributor

Posts: 38

Joined: Thu Apr 13, 2017 8:46 pm

Post Thu Oct 26, 2017 12:19 am

Re: Verity Cen-se?

Hi El-ad,

I am not sure how to quote you, so I'll just keep greeting you as such. :)

Maybe I'm confused, but I was using that benchmark to assess aim (1) unbiassed singlet filtering, not double-singlet separation. When Bagwell refers to complexity or composition in the sample during that point of the talk, I do not believe he is referring to doublet-singlet separation.

Matthew
<<

AdeebR

Grand master

Posts: 169

Joined: Thu Mar 13, 2014 5:58 pm

Location: NYC

Post Thu Oct 26, 2017 12:50 am

Re: Verity Cen-se?

kunicki wrote:Hi El-ad,
I think this difference in DNA-Ir signal between populations came up earlier in the forum in a separate thread.


viewtopic.php?f=3&t=374&hilit=doublet
Adeeb Rahman
Icahn School of Medicine at Mount Sinai, NYC
<<

mleipold

Guru

Posts: 5796

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Fri Oct 27, 2017 12:35 am

Re: Verity Cen-se?

El-ad:

"But how can we use this as a benchmark? As far as I know there is no quantitative evidence that high DNA-Ir has anything to do with doublets! We're just following Bendall et al's 2013 work."

If you gate progressively tighter toward Ir-low, you get rid of progressively more doublets like CD3+CD20+, CD3+CD14+, and CD14+CD20+ cells.

Granted, a lot of this is the fact that you get rid of more and more EL-long cells as you do that.


You can see this in the attached file (Helios data): this is a healthy control PBMC sample where I gated in various ways on Ir and not EL (page 1), or Ir-broad and then EL-low vs EL-high vs EL-all (page 2), and then in each case gates on Live cells before further marker-based gating.

The middle plots are the doublet plots mentioned above. The very bottom plots are the same Live cells, but gated as CD33 vs CD14 to watch how the Monocytes respond to the various gating schemes.


Granted, this is one sample, but is consistent with all the other samples I've seen. In short: EL-long is a stronger determinant of doublets than just Ir-bright. But the EL-long are where most of the doublets are, and almost all of them *are* Ir-bright. But there's also the Ir-bright-EL-low population that you don't want to miss, as that's where some of your Monocytes are!


Mike
Attachments
EL and Ir on Doublets.pdf
(826.88 KiB) Downloaded 377 times
<<

dtelad11

Master

Posts: 129

Joined: Mon Oct 31, 2016 6:26 pm

Post Fri Oct 27, 2017 12:52 am

Re: Verity Cen-se?

mleipold wrote:If you gate progressively tighter toward Ir-low, you get rid of progressively more doublets like CD3+CD20+, CD3+CD14+, and CD14+CD20+ cells.


To clarify, I am not saying that the statement regarding DNA-Ir is *wrong*. In the past I used DNA/Event length as the first step of data processing as well. And I fully trust your judgement on this, I am sure that you have worked on thousands or even tens of thousands of samples by now, I trust your expertise.

My point is that there are no quantitative benchmarks that show this. Nobody has done the controlled experiment. Right now it's all anecdotal. If I have five methods that pre-process mass cytometry data, I have no feasible, reproducible way of quantifying which is better and which is worse.
Previous

Return to CyTOF data analysis

Who is online

Users browsing this forum: No registered users and 10 guests