separating type (lineage) and state (functional) markers
Hi all,
First, I wanted to say that Lukas' diffcyt framework has been updated ..
Preprint: https://goo.gl/mED5pa
Software: http://bioconductor.org/packages/releas ... ffcyt.html
Vignette: http://bioconductor.org/packages/releas ... kflow.html
Second, I wanted to solicit some feedback in relation to some feedback we received from reviews on the diffcyt paper. I posted this recently on Twitter (https://twitter.com/markrobinsonca/stat ... 5390495744), but I thought that this mailing list might be a more relevant place to ask my question. Below is more-or-less a cut-and-paste of it:
An interesting thread came through pretty strong from the reviewers .. and (ourselves) have gone back-and-forth on this discussion many times, so I thought I would spell it out a bit here ..
So, basically what we do in our cytometry "differential discovery" analyses (think: 1000s of cells for each sample, 20-50 markers measured, multiple samples across experimental conditions) is split the markers into "type" (those that we cluster on) and ..
"state" (not clustered on, but we are interested in type-specific changes therein). So, we frame the statistical problem as i) differences in abundance of cell types/clusters .. or ii) given a cell "type", changes in "state". The reviewers rightly challenged this ..
saying "I like the way that the authors propose to model [type] and [state] markers differently, although it’s not clear that this should always be done" and "there is no consensus in the field regarding the difference between ‘cell-type’ and ‘cell-state’" .. we totally agree!
But, we still think that a method should offer the capability to separate type/state (many don't). There are cases when it's clear. Re: sensitivity, the method should detect interesting differences no matter how the dichotomy is set and we think it helps interpretation.
Any opinions about this?
Thanks in advance and best regards, Mark
First, I wanted to say that Lukas' diffcyt framework has been updated ..
Preprint: https://goo.gl/mED5pa
Software: http://bioconductor.org/packages/releas ... ffcyt.html
Vignette: http://bioconductor.org/packages/releas ... kflow.html
Second, I wanted to solicit some feedback in relation to some feedback we received from reviews on the diffcyt paper. I posted this recently on Twitter (https://twitter.com/markrobinsonca/stat ... 5390495744), but I thought that this mailing list might be a more relevant place to ask my question. Below is more-or-less a cut-and-paste of it:
An interesting thread came through pretty strong from the reviewers .. and (ourselves) have gone back-and-forth on this discussion many times, so I thought I would spell it out a bit here ..
So, basically what we do in our cytometry "differential discovery" analyses (think: 1000s of cells for each sample, 20-50 markers measured, multiple samples across experimental conditions) is split the markers into "type" (those that we cluster on) and ..
"state" (not clustered on, but we are interested in type-specific changes therein). So, we frame the statistical problem as i) differences in abundance of cell types/clusters .. or ii) given a cell "type", changes in "state". The reviewers rightly challenged this ..
saying "I like the way that the authors propose to model [type] and [state] markers differently, although it’s not clear that this should always be done" and "there is no consensus in the field regarding the difference between ‘cell-type’ and ‘cell-state’" .. we totally agree!
But, we still think that a method should offer the capability to separate type/state (many don't). There are cases when it's clear. Re: sensitivity, the method should detect interesting differences no matter how the dichotomy is set and we think it helps interpretation.
Any opinions about this?
Thanks in advance and best regards, Mark