Addressing the inter-channel problem
Peter McLean
A partial review of our last conversation and my thoughts on what we can do.
I will refer to Jake’s summary of the flowcyt measurement process:
Asc = Pc Ns Xs(fc) [ <integral> Es(f) · Tc(f) df ] αc
Where:
P = laser power
N = number of species (for a given bead peak population)
X(fc) = species excitation, for a given channel
E(f) = species emission (normalized, wavelength agnostic… mostly)
T(f) = light-path transmission
α = detector amplification
The TLDR of the `units.pdf` in the sbsc `publications` directory is that P, T, and α are constant for a given instrument configuration, N is what we’re ultimately trying to measure, and the rest are features of the measured species.
P, T and α are effectively scalars that can be controlled with any fluorescent reference particle, including the RCPs (rainbow calibration particles).
We have reasonable data for the emission spectra for all the things we are measuring, so it is pretty straightforward to scale measurements made with an arbitrary band-pass to that which was used for calibrating the RCPs (e.g., FITC, PE, etc.) by taking the ratio of the integrals for each band-pass. As far as I can tell, this will ONLY work when the same excitation source is used. This means it’s great for converting measurements from a (hypothetical) 530/50 filterset to a 525/20 filterset—so long as the excitation light is 488 nm for both.
That leaves us with X(fc) still unaccounted for. This term is absolutely critical for comparing measurements made among different excitation lines, which is a primary goal for this project. In short, the problem here is that even though the emission profile of a fluorescent molecule does not change much with different excitation wavelengths (even RCPs, mostly), it’s total photon output absolutely does.
_____
At this point, I submit the idea that X(fc), as currently defined, is really just the quantum yield (QY) for the fluorescent species being measured. Because QY is linear with respect to concentration, it basically states that for X number of absorbed photons of a given wavelength, X*QY photons will be emitted by fluorescent decay. In our case, X is pretty much just a scaled value of P, and I don’t think we need to worry about differences in the incidence of photon absorption between dyes and fluorescent proteins. We’ll save that for version 2.0.
How would this work? Well, in addition to the emission spectra for all measured species, we would also need a QY value for our excitation wavelengths; e.g., QY values for the RCPs at 405, 488, 543, 561, 633 nm. QY values abound for fluorescent dyes and proteins, so we’re already covered there.
Importantly, and in contrast to how I’d been thinking about this problem for a while, this approach would mean that calibration “path” for fluorescent proteins would likely be:
Raw-signal-values > nearest-calibrated-RCP-channel > common-scale
The main reason for this is that we CAN establish high-confidence transform functions between different dyes like FITC and PE (even using different excitation wavelengths) using some carefully prepared dilutions and a calibrated fluorometer. NOTE: this will NOT provide a quantitative metric of how many fluorescent proteins are in the sample… it will tell you how bright your sample is in units of whatever calibrated solution you’d like—FITC, PE, CY5, etc.—and will allow you to compare signal brightness across excitation lines.
It seems reasonable that you could get to number-of-proteins (or at least close) if you separately produced something like a FITCxGFP curve, but that’s a whole different set of problems. I also imagine that this might NOT end up with a 1:1 relationship for two different fluorescent proteins in cases where you might expect such (i.e. dual-expression controls for inter-fluor calibration). That said, it does serve the purpose of getting all channels onto the same scale in a very controlled, calibrated manner and could be used in all the same applications.
Thoughts? Suggestions? Is this all completely wrong? Tune in Monday, 1PM PT, 4PM ET.
Peter McLean
I have added a `pfm` folder into the analysis directory of the DeLateur data. In it you will find an HTML notebook in which i show the my latest iteration of measurement conversion. I think it is pretty well annotated, so I won't describe it again in this post. That said, here's your teaser...
...it's very encouraging, at the least. In that folder you'll also find the experiment and reference tables I used to process the data, and even a summary xlsx that you can use to plot the results yourself. NOTE: I did use a modified version of cytoflow, so you won't be able to run the .ipynb on your own.
~ Peter
Brian Teague
>
> I have added a `pfm` folder into the analysis directory of the DeLateur
> data. In it you will find an HTML notebook in which i show the my
> latest iteration of measurement conversion. I think it is pretty well
> annotated, so I won't describe it again in this post. That said, here's
> your teaser...
opening the Jupyter notebook, here are the takeaway plots.
--
Brian Teague
tea...@mit.edu
Weiss Group, Synthetic Biology Center @ MIT
