FAQ  •  Register  •  Login

Auto-gating Gaussian and/or cells?

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 :-)
<<

mleipold

Guru

Posts: 5792

Joined: Fri Nov 01, 2013 5:30 pm

Location: Stanford HIMC, CA, USA

Post Sat Dec 07, 2019 4:29 pm

Re: Auto-gating Gaussian and/or cells?

There's also a general issue of using hard cutoffs. I had a sample a couple years ago where there was a La139 contamination. It was so bad that the left (138) and right (140) spillovers were at least approaching 1e2. When I'm back to the lab next week, I can try to dig it up (it's a teaching slide now...).

Similarly, there are a few labs who use Ce as another antibody channel. It doesn't interfere with the normalization bc the cells are not positive for all 5 Norm channels.

But that would warp your Ce140-only cutoff.
<<

jimbomahoney

Master

Posts: 83

Joined: Wed Feb 27, 2019 11:21 am

Post Sun Dec 08, 2019 2:41 pm

Re: Auto-gating Gaussian and/or cells?

Thanks Mike and El-ad.

Mike - I’ve seen your post and screenshots of the issue you mention, so no need to dig it up.

Given these issues, I think I’ve come up with a nice solution.

I’ll also make the bead gating optional and dependent upon at least one bead channel being present.

Seems to be working nicely and I agree that hard cut-offs are lazy!

EDIT - I’ll post back when it’s on GitHub.
<<

jimbomahoney

Master

Posts: 83

Joined: Wed Feb 27, 2019 11:21 am

Post Mon Dec 09, 2019 7:26 pm

Re: Auto-gating Gaussian and/or cells?

OK, after much testing and tweaking, I think I've come up with a method to gate the beads (i.e. remove them).

The new version on GitHub has two significant changes:

1) Removing Beads is now optional.
2) If bead removal is enabled (default), the script will examine each bead channel.
3) Whichever channel has the clearest separation between the cells (low intensity) and beads (high intensity), it will use that Channel to set the cut-off appropriately.

This should work even when markers are used in one or more of the bead channels.

It works on all the datasets I have, some of which have markers in most of the bead channels.

However, I can see a possibility of the bead removal failing, or not working properly, if ALL the bead channels are "noisy" as a result of being used for cell markers. Hence why the bead removal is now optional.

UPDATE 3rd March 2020 - OK, the new version now also asks the user which bead channel(s) they want to use. Ideally, these would not have been used for markers.
Last edited by jimbomahoney on Tue Mar 03, 2020 3:16 pm, edited 1 time in total.
<<

AdeebR

Grand master

Posts: 169

Joined: Thu Mar 13, 2014 5:58 pm

Location: NYC

Post Wed Dec 11, 2019 10:19 pm

Re: Auto-gating Gaussian and/or cells?

In identifying the beads, I think it's helpful to consider the fact that the beads are in fact uniformly positive for all the bead channels. So rather than relying on any single channel for bead identification, why not just use all of them (i.e., even if you have noisy background in multiple channels, it's unlikely that any given noise event will be equally positive in Ce140, Eu153 and Lu175, whereas you can be confident that the beads will be). So even in samples where we've had high noise in several channels, the beads typically still resolve clearly by tSNE.
Adeeb Rahman
Icahn School of Medicine at Mount Sinai, NYC
<<

AdeebR

Grand master

Posts: 169

Joined: Thu Mar 13, 2014 5:58 pm

Location: NYC

Post Wed Dec 11, 2019 10:19 pm

Re: Auto-gating Gaussian and/or cells?

In identifying the beads, I think it's helpful to consider the fact that the beads are in fact uniformly positive for all the bead channels. So rather than relying on any single channel for bead identification, why not just use all of them (i.e., even if you have noisy background in multiple channels, it's unlikely that any given noise event will be equally positive in Ce140, Eu153 and Lu175, whereas you can be confident that the beads will be). So even in samples where we've had high noise in several channels, the beads typically still resolve clearly by tSNE.
Adeeb Rahman
Icahn School of Medicine at Mount Sinai, NYC
<<

AdeebR

Grand master

Posts: 169

Joined: Thu Mar 13, 2014 5:58 pm

Location: NYC

Post Wed Dec 11, 2019 10:21 pm

Re: Auto-gating Gaussian and/or cells?

In identifying the beads, I think it's helpful to consider the fact that the beads are uniformly positive for all the bead channels. So rather than picking any single channel as the "best" for bead identification you could just use all of them (i.e., even if you have noisy background in multiple channels, it's unlikely that any given noise event will be equally positive in Ce140, Eu153 and Lu175, whereas you can be confident that the beads will be). So even in samples where we've had high noise in several channels, the beads typically still resolve clearly by tSNE.
Adeeb Rahman
Icahn School of Medicine at Mount Sinai, NYC
<<

jimbomahoney

Master

Posts: 83

Joined: Wed Feb 27, 2019 11:21 am

Post Thu Mar 05, 2020 11:41 am

Re: Auto-gating Gaussian and/or cells?

Significant update just released (v 1.0 beta).

https://github.com/JimboMahoney/cytofclean

Improvements:

1) User can now define which bead channel(s) to use for bead removal. This helps cytofclean to "know" which do not contain other markers.
2) All parameters (except Event Length) are now arcsinh transformed (cofactor 5) to perform the gating. Raw data is NOT affected, only a copy of the data is used to determine the gates.
3) Hard Event cutoffs applied (less than 10 and greater than 50 pushes are excluded). This only kicks in with "bad" datasets. The vast majority fall in the ~12 - 30 range.

The results are much better than previous versions (less aggressive) and plots are similar units / scale to cytobank.
<<

sgranjeaud

Master

Posts: 123

Joined: Wed Dec 21, 2016 9:22 pm

Location: Marseille, France

Post Thu Mar 19, 2020 2:39 pm

Re: Auto-gating Gaussian and/or cells?

Hi James,
Interesting work. I tested your package which does a good job.
Why don't you do filtering on DNA1 or DNA2 as shown by Bagwell-2019 in Fig. 4?
Best,
Samuel
Last edited by sgranjeaud on Sat Mar 21, 2020 8:54 pm, edited 1 time in total.
<<

mdmanurung

Participant

Posts: 2

Joined: Thu Nov 23, 2017 11:09 am

Post Sat Mar 21, 2020 10:18 am

Re: Auto-gating Gaussian and/or cells?

I have tested this package and I have to say that you have done such an amazing job!! I would also make a feature request for gating DNA :D

I think you have spotted a great problem or bottleneck that can be automatized.
<<

jimbomahoney

Master

Posts: 83

Joined: Wed Feb 27, 2019 11:21 am

Post Mon Mar 23, 2020 8:03 am

Re: Auto-gating Gaussian and/or cells?

Thanks both!

I'll add that to the list of things to do!
Previous

Return to CyTOF data analysis

Who is online

Users browsing this forum: No registered users and 6 guests