Minian Decay Times

[Marek]

Dear all,

we were running experiments with the V4 miniscopes and started to analyze them using minian. We are quite happy with the preprocessing steps and it's also cool how the pipeline leads one through the parameter exploration / selection.

However, we see largely prolonged decay times in the computed calcium traces C, if compared to the raw traces. I know there is no single parameter to drive for this, but I have played around with the others quite extensivley and couldnt find settings that have a major impact on the decay times. Caiman has a parameter to play with for decay times, and without tweaking any of the native settings, the deconvolved traces look much better on the same data set already (the fov seems to be reversed as well).

Did anybody encounter similar phenomenons on decay times and/or has tips how to deal with it? For example, are there parameters in the source code one can change to tweak that?

 

[Daniel Aharoni]

Hi Marek,

You bring up good points and often times, this deconvolution step is the most tricky and can lead to the biggest issues in downstream analysis when not done properly. Minian, along with most other 1-photon CA+2 processing pipelines, attempt to automate the process of finding the proper time-constants for the deconvolution algorithm. This, unfortunately, often leads to less than optimal output. 

What my group does is changes which deconvolution equation is used (we usually use the double exponential) and then have a workflow for determining the 2 time constants that best fit our data. We apply the same 2 constants to all cells/traces within a give recording. 

The details of how specifically we do this are probably a bit too in-depth for this form but I will see if we can write something up and share it here.

[Marek]

Hi Daniel,

Many thanks for your answer! I'm also using a double exponential equation, but it would be great to know a bit more about how you determine the time constants! A write up about that would be super informativ, I guess also for other starters. However, some code snippets would help as well if you are short in time!

[Nate Klett]

This issue is discussed some here in the #3 Hands-On Analysis Tutorial from the 2021 Miniscope Workshop. In the video, this is discussed around the 1:28 time point. 

Phil says that the fit of the decay is generated from an auto-regressive model that is estimated automatically based has a couple of parameters, that cannot be adjusted. However, he indicates that these parameters, may be able to be "hacked," as Daniel does. 

There is however a discussion that adjusting the value of add_lag, may influence the fit of the transient. Add_lag does not change the fit, but just how many data points are used to perform the fit.

[Nate Klett]

Can someone explain to me what Minian's c0 matrix is? Are these the modelled fits of the simulated calcium transients for each cell/unit_ID?