Units without spectra
4 views
Skip to first unread message
jakebeal
Nov 21, 2016, 3:00:54 PM11/21/16
to SBSC Flow Cytometry Working Group
For discussion at today's meeting: I've satisfied myself that we can convert between channels for rainbow beads without bead spectra as long as we have spectra for the ERF. My reasoning is spelled out in Figure 3 and Section 5, starting on page 4.
Thanks,
-Jake
Peter McLean
Nov 21, 2016, 8:14:12 PM11/21/16
to SBSC Flow Cytometry Working Group
The followup is awesome. Straight to my questions/comments:
- I'm concerned about the unit descriptions in the main measurement process equation. the amplifier is controlled in units of Volts, but the scaling effect on the data is not linear with respect to the applied voltage. I'm also super suspicious of putting any units on this because the PMT is acting as a photon-packet counter. The signal magnitude of the cascade output is proportional to the photon-density of the input, but it seems to me that the unit shouldn't change... which is something along the lines of photons. These detectors cycle at some Hz so that as cells pass through the collection plane, you obtain 1e2-1e6 (instrument and cell size depending) measurements that we use to calculate area, width, and height of the event response. In the end—if you pretend that your baseline restoration circuitry is perfect, and your trigger thresholds are perfect, and your detector efficiency is perfect, you could divide out the PMT amplification factor and get the exact number of photons emitted by the cell as it passed through the excitation line. In reality, the output is a (unit-less) signal value that is proportional to the number of photons emitted... and empirically, it appears to be reasonably linear. TLDR: giving anything units of watts or "energy" makes me nervous... it seems like an unnecessary step into the physics world that introduces more problems than it solves.
- I believe I've employed all the same logic in my write-up, though it is written from a different perspective. I too came to the same conclusion that you wrote into section 5... but I think it's worth restating that this only works if the excitation line and light-path (filterset) of YOUR measurements are the SAME as those used for calibration of the unit scale. For example, Spherotech reports that they calibrated their MEFL channel with a 530/40 filter (presumably the 488 nm excitation line). If you have that same filterset, you're good to go. If you used a 530/30 filter—like I did—then you start to break the assumptions for this equation. Now, I think it's a fair argument to say that these two filters may not have significant differences, but that won't always be true. Importantly, the fix for this is to rescale the data to reflect the REFERENCE band-pass—so, multiply your data by a reference factor based on the reporter emission spectrum to represent the signal you WOULD have measured had you used a 530/40 filter instead of a 530/30.
If you're talking about using different excitation lines, the problem gets worse, for reasons we've discussed at length before—beads and reporters don't respond equally to different excitation wavelengths. In this scenario you'd need to accommodate those differences if you want to calibrate an arbitrary channel to a reference channel of a different excitation line. As of today, I'm of the belief that the most reasonable approach to solve this would be to convert your reporter measurements to the nearest, same-excitation unit scale FIRST, using your described methodology. So, GFP to MEFL, mCh to MEPTR, etc. The beads themselves give you the rationale to make those equivalent and transform between those scales... but again, only after you've appropriately converted your data to the same "conditions" that were used for the reference particle calibrations.
IMO, the biggest values of this work are A) to enable inter-channel comparability and justify dual-reporter quantitation using an existing, stable, reference artifact, and B) to place all measurements on a common scale which enables experiment troubleshooting (anomaly detection), improves interpretability of experimental results (no more floating baseline/negative control value), and facilitates the reproduciblity of cell characterization. It seems to me that we're circling all this pretty well... just a little more pushing and it'll get there.
I'll try to update my writing a bit to reflect some of this, and hope you find some of it useful!
Peter
- I'm concerned about the unit descriptions in the main measurement process equation. the amplifier is controlled in units of Volts, but the scaling effect on the data is not linear with respect to the applied voltage. I'm also super suspicious of putting any units on this because the PMT is acting as a photon-packet counter. The signal magnitude of the cascade output is proportional to the photon-density of the input, but it seems to me that the unit shouldn't change... which is something along the lines of photons. These detectors cycle at some Hz so that as cells pass through the collection plane, you obtain 1e2-1e6 (instrument and cell size depending) measurements that we use to calculate area, width, and height of the event response. In the end—if you pretend that your baseline restoration circuitry is perfect, and your trigger thresholds are perfect, and your detector efficiency is perfect, you could divide out the PMT amplification factor and get the exact number of photons emitted by the cell as it passed through the excitation line. In reality, the output is a (unit-less) signal value that is proportional to the number of photons emitted... and empirically, it appears to be reasonably linear. TLDR: giving anything units of watts or "energy" makes me nervous... it seems like an unnecessary step into the physics world that introduces more problems than it solves.
- I believe I've employed all the same logic in my write-up, though it is written from a different perspective. I too came to the same conclusion that you wrote into section 5... but I think it's worth restating that this only works if the excitation line and light-path (filterset) of YOUR measurements are the SAME as those used for calibration of the unit scale. For example, Spherotech reports that they calibrated their MEFL channel with a 530/40 filter (presumably the 488 nm excitation line). If you have that same filterset, you're good to go. If you used a 530/30 filter—like I did—then you start to break the assumptions for this equation. Now, I think it's a fair argument to say that these two filters may not have significant differences, but that won't always be true. Importantly, the fix for this is to rescale the data to reflect the REFERENCE band-pass—so, multiply your data by a reference factor based on the reporter emission spectrum to represent the signal you WOULD have measured had you used a 530/40 filter instead of a 530/30.
If you're talking about using different excitation lines, the problem gets worse, for reasons we've discussed at length before—beads and reporters don't respond equally to different excitation wavelengths. In this scenario you'd need to accommodate those differences if you want to calibrate an arbitrary channel to a reference channel of a different excitation line. As of today, I'm of the belief that the most reasonable approach to solve this would be to convert your reporter measurements to the nearest, same-excitation unit scale FIRST, using your described methodology. So, GFP to MEFL, mCh to MEPTR, etc. The beads themselves give you the rationale to make those equivalent and transform between those scales... but again, only after you've appropriately converted your data to the same "conditions" that were used for the reference particle calibrations.
IMO, the biggest values of this work are A) to enable inter-channel comparability and justify dual-reporter quantitation using an existing, stable, reference artifact, and B) to place all measurements on a common scale which enables experiment troubleshooting (anomaly detection), improves interpretability of experimental results (no more floating baseline/negative control value), and facilitates the reproduciblity of cell characterization. It seems to me that we're circling all this pretty well... just a little more pushing and it'll get there.
I'll try to update my writing a bit to reflect some of this, and hope you find some of it useful!
Peter
Reply all
Reply to author
Forward
0 new messages