mouseland / suite2p Goto Github PK

Also, add hdf5 as possible input to pipeline.

Select multiple ROIs, for example with CTRL + CLICK or with a lasso.

Show all their traces (but none of the neuropil traces maybe?) and allow users to discard all ROIs simultaneously. This would be particularly convenient with the lasso function.

...during extraction by looking at pixel neighbors. This would remove the need to keep ROIs local, but would still require them to expand via connected regions. Can this work?

@carsen-stringer where do I find and how do I call the classifier to produce the probability based on stat?

will be fixed in next release
pyqtgraph/pyqtgraph#722

for now, please export to *.svg

Sometimes the ROIs are packed so tight that there are few neuropil pixels left. This is not too bad, as it often happens when there is not a lot of neuropil contamination, perhaps as a consequence. We should still have a way to treat such dense recordings. There are a few options, in order of difficulty:

1. have explicit user-defined flag for processing a dense recording
2. detect dense recordings automatically and switch to dense mode automatically
3. integrate dense extraction seamlessly into the algorithm somehow

Do this in the first iteration, same place where we define a threshold.

Also, allow different path on a fast SSD

This one should come as default with Suite2p.

How should we support non-localized ROIs, like long dendrites, or axons with multiple boutons?

One argument is they should be split up into pieces, since different segments might respond differently. We can then have a posthoc step to merge them back, similar to this Matlab implementation. The posthoc step might require proximity and high correlation, like issue #28 .

We could use the GUI protocol for merging ROIs, so that users can also un-merge them in the GUI as proposed in issue #17 . We can still use the binary tree procedure to merge sequentially, since this will at most double the number of ROIs, and these types of fields of view will be sparse anyway.

And start making a database and a default classifier.

Should be converted to binary right away.

Add merging ability to the GUI. Multi-neuron selection will be provided by issue #16.

If I load a recording in the GUI, and then I load another one, sometimes I get errors like this:

H:/DATA/Allen/rec575766607/suite2p/plane0/stat.npy
no manual labels found (iscell.npy)
===== 2018.08.06 10:08:57 =====
Traceback (most recent call last):
File "D:\Github\suite2p\suite2p\gui2p.py", line 440, in load_dialog
self.load_proc()
File "D:\Github\suite2p\suite2p\gui2p.py", line 477, in load_proc
self.make_masks_and_buttons()
File "D:\Github\suite2p\suite2p\gui2p.py", line 284, in make_masks_and_buttons
fig.init_masks(self)
File "D:\Github\suite2p\suite2p\fig.py", line 53, in init_masks
i = int(1-iscell[n])
IndexError: index 384 is out of bounds for axis 0 with size 346

And change them to say cells not cells, and change color when pressed. Allow both to be pressed, but don't allow both to be unpressed.

and @marius10p add to pipeline

This could reproduce the Matlab version.

Some people want this in order to know how many cells are active roughly.

For example

1. raster plot of fluorescence/spikes.
2. sorted raster plot with neurons and/or timepoints sorted by PC1 or by a 1D embedding.
3. lasso groups of neurons and do the visualizations just for them

The diffuse neuropil ROIs that make it through. Could also do this iteratively and allow the other ROIs to adjust.