Converted deconvolution data into spike train (binary point)

[zhouyi0812]

Hello!

After I download the deconvolution trace from estimates.S, I got deconvoluted time series data. However, I want the spike train (point data) so I use another package to select the peak time point and convert it to 1 and the rest of the time point is converted to 0. I wonder are you also doing this way? If there is another way to handle it in CaImAn that I missed.

Thank you!!!

[kushalkolar]

Do you just want to do 0 - 1 normalization? The easiest thing is to just use tslearn min max scaler: https://tslearn.readthedocs.io/en/stable/gen_modules/preprocessing/tslearn.preprocessing.TimeSeriesScalerMinMax.html

[zhouyi0812]

I think it is not normalized to (0,1). For example, the deconvolution trace is 0, 0, 4.3, 5.4, 6.6, 0, 0 and 6.6 is the peak and my final data is 0, 0, 0, 0, 1, 0, 0

[kushalkolar]

So you just want an array where the max is 1 and rest 0? Not sure what you're really trying to do but:

a[a < a.max()] = 0
a[a > 0] = 1

[zhouyi0812]

You're not doing this way? So how you plot this kind of raster plot?

[zhouyi0812]

[zhouyi0812]

[kushalkolar]

Converting S to spike count isn't as trivial as max(S). See this discussion: #499

[EricThomson]

I interpreted it as you want a raster plot like thing. This is not currently part of the estimates object, which gives you an estimate of spike count in estimates.S on the same time scale as the sampling rate of your experiment. It is not possible, afaik to get a finer time scale. This is using OASIS (implemented in constrained_oasisAR2() function), and things get more complex with other algorithms like foopsi. OASIS works really well so I stick with that.

Though you didn't say what type of data you are analyzing iirc you are looking at 1p. If 1p data I would be wary of deconvolution results tbh we don't have any ground truth data. I'd probably stick with analyzing traces if using CNMFE.

[zhouyi0812]

So basically I need to get the trace from here: https://github.com/EricThomson/yacare/blob/main/normalize_1p_traces.ipynb

Then find the peak and plot the raster plot. Thank you!

[kushalkolar]

It still wouldn't make much sense to set the max as 1 and everything else as 0 if you want to understand the activity of that neuron. You can plot a heatmap of dfof or zscored activity using seaborn, for example with C:

import seaborn as sns
from tslearn.preprocessing import TimeSeriesScalerMeanVariance

zscore = TimeSeriesScalerMeanVariance().fit_transform(cnmf_obj.estimates.C)

sns.heatmap(zscore)

[zhouyi0812]

Thank you very much! I want to use that because I saw some downstream analysis from this tool: https://elephant.readthedocs.io/en/stable/

Also, may I ask have you used multivariate granger causality to analyze and cluster the trace? Thank you!

[kushalkolar]

Let me try to clear one misconception, you can't really deconvolve calcium traces into ephys waveforms. You can get rough estimates of spike rate. This paper has a good into that discusses it: https://www.nature.com/articles/s41593-021-00895-5

…

On Wed, Oct 18, 2023, 12:08 Yi Zhou ***@***.***> wrote: Thank you very much! I want to use that because I saw some downstream analysis from this tool: https://elephant.readthedocs.io/en/stable/ Also, may I ask have you used multivariate granger causality to analyze and cluster the trace? Thank you! — Reply to this email directly, view it on GitHub <#1195 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACHXXRFRLBGBVRYOSFOLBTTX775HJAVCNFSM6AAAAAA6ER6EOWVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM3TGMJXGM2TS> . You are receiving this because you commented.Message ID: ***@***.*** com>

[zhouyi0812]

Thank you very much for your clarification!
While I trying to use the code above, it seems I need to save hdf5 file and then convert to cnmf.object
I am confused about this 'batch_dataframe_file.pickle' and replace with my hdf5 file. But it says AttributeError: 'CNMF' object has no attribute 'iloc'. Could you give some guidance? Thank you!

from mesmerize_core import load_batch

df = load_batch("/path/to/batch_dataframe_file.pickle")

assume the 0th index is a cnmf item

cnmf_obj = df.iloc[0].cnmf.get_output()

see some estimates

print(cnmf_obj.estimates.C)
print(cnmf_obj.estimates.f)

[EricThomson]

@zhouyi0812 it sounds like this last question is a mesmerize question, have you resolved it?