data processing issues for recent MESO2 data in LearningmFISHTask1A project code

[matchings]

I have been trying to analyze data from a Gad2 mouse (mouse_id: 603892) in the LearningmFISHTask1A project code and it appears that many of the experiments have zero, or very few, segmented ROIs, despite there being clear cells in the max intensity and average projections. This was not an issue for a different mouse (mouse_id: 582466) from the same cre line, on the same microscope, in LearningmFISHDevelopment project code last December.

I have compiled the decrosstalk QC plots from these two mice into a folder here:

/allen/programs/mindscope/workgroups/learning/ophys/qc_plots/session_plots/decrosstalk_plots

In these plots, you can see which experiments / sessions have segmented ROIs and which do not.

affected ophys_session_ids from mouse 603892 include: [1154000392, 1156589145, 1155380558, 1155606631, 1156842547,
1153776022, 1154460714, 1154245678, 1155053105, 1155848487, 1156039109] (although the later sessions do look slightly better).

@wbwakeman has tried reprocessing the more recent problematic data data with a "mode" of 6, which is typically good for inhibitory lines (Gad2 is a pan-inhibitory line), and this may have helped somewhat but did not solve the problem. Wayne also tried changing the setting from 14-bit to 16-bit which has previously been identified as a possible need for MESO2 data, but that did not solve the problem either.

@danielsf tested suite2P motion correction on the data from mouse 603892 and the results were somewhat better but there are still visibly unsegmented ROIs.

/allen/aibs/informatics/danielsf/mfish_learning/segmentation_220216/plots

This led me to believe that it could be a data quality issue, but what I hear from the ophys team is that the data quality on this mouse has been excellent, and the motion corrected movies look perfectly fine to me. Perhaps the different SNR of this new cre line is throwing things off as it is GCaMP7s which has a higher SNR than the GCaMP6f mice we have used in the past, but that does not explain why things worked ok on a previous mouse of this type but not the more recent mouse.

I am happy to help troubleshoot this issue, but I will not have time to help until April due to other deadlines. Perhaps @nataliaorlova or others on the ophys team may be able to help.

[wbwakeman]

A part of this story is that ROIs are segmented, but they are marked invalid. For various reasons. In fact, here is the breakdown. Recall that one ROI can have multiple reasons for being marked 'invalid'.
label count
bad_shape 2
boundary 5
decrosstalk_invalid_raw 12
decrosstalk_invalid_raw_active 14
decrosstalk_invalid_unmixed 8
low_signal 19
motion_border 8
small_size 3