linear regression to remove motion artifects

[Feikai-Lin]

Hi Thomas,

Thank you for sharing this pipeline. It is very clear and helpful! When I applied linear regression with detrended isosbestic channel to remove motion artifacts in the detrended signal, as you did in the notebook, I noticed that the slopes are sometimes negative. It's surprising, and even counterintuitive, to see that when artifacts are larger in isosbestic, they are smaller in the signal. I'm not sure why, and I wonder if you have seen something similar before? Many thanks!

Best, Feikai

[ThomasAkam]

Hi @Feikai-Lin,

You can get bleed-through of the physiological signal to the isosbestic channel if the excitation wavelength for the isosbestic channel is not exactly at the isosbestic point of the sensor, so the sensor still has some sensitivity to the physiological change being measured. This bleed-through can be negative if e.g. an increase in calcium (if using GCaMP) reduces the fluorescence at that excitation wavelength. I have observed such negative bleedthrough with GCaMP6 excited at 405nm. For more discussion of these issues see the photometry primer article we published last year, and particularly figure 2. As shown in figure 2 you can often assess if there is significant signal bleed-through into the control channel by looking at the average control channel signal aligned on an event that is know to produce a large change in the physiological signal.

If there is significant bleed-through of the physiological signal into the control channel it is not possible to use the control channel to correct for movement artifacts. Due to this issue of signal bleed-through into isosbestic channels, and the insensitivity of isosbestic signals to movement artifacts due to the large contribution of bulk autofluorescence, we instead use a seperate red flurophore for movement control in our experiments.

best,

Thomas

[Feikai-Lin]

Hi Thomas,

Thanks a lot! I am also using GCaMP6f, and it seems to have a small bleed-through.

Best, Feikai