search

jovo / oopsi

up-to-date code for model-based inference of spike trains from calcium imaging
Apache License 2.0
24 stars 9 forks source link

What does OOPSI stand for? #5

Closed guidomeijer closed 8 years ago

guidomeijer commented 8 years ago

I can't seem to find out and the mystery is killing me

jovo commented 8 years ago

Optimal Optical Physiological Spike Inference

;-)

i think its in my thesis?

On Thu, Sep 29, 2016 at 8:32 AM, Guido Meijer notifications@github.com wrote:

I can't seem to find out and the mystery is killing me

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/jovo/oopsi/issues/5, or mute the thread https://github.com/notifications/unsubscribe-auth/AACjcoB7-TVy5ksSIb7kGNOOZA_g0Ww1ks5qu6_CgaJpZM4KJ75q .

the glass is all full: half water, half air. neurodata.io, jovo calendar https://calendar.google.com/calendar/embed?src=joshuav%40gmail.com&ctz=America/New_York

guidomeijer commented 8 years ago

Thanks! I was looking for it in the 2010 paper but couldn't find it.

The algorithm works really well for me! I've run fast-OOPSI on the publicly available dataset from the Chen lab with GCaMP6f recordings and ground truth spike trains from patch clamp recordings. I've used different sets of parameters to find the ones with the highest correlation between inferred and 'real' spike trains. I've found the best performance is with a tau of 0.8 and and estimated firing rate of 0.04. I've set the standard deviation of the measurement noise to the standard deviation over the lowest 50% of points of the dF/F trace.

If I then run fast-OOPSI on the dataset the correlation between the true and inferred spike train is very good (similar to Theis et al 2016) and there is a nice linear relationship between inferred spike train and actual number of spikes in a 100 ms timebin. I've included my findings as a figure, so you can check it out.

Thanks for making this code publicly available! spikeinferenceoopsi-01

jovo commented 8 years ago

wow, the data look really great, thanks so much for the feedback!

On Thu, Sep 29, 2016 at 9:57 AM, Guido Meijer notifications@github.com wrote:

Thanks! I was looking for it in the 2010 paper but couldn't find it.

The algorithm works really well for me! I've run fast-OOPSI on the publicly available dataset from the Chen lab with GCaMP6f recordings and ground truth spike trains from patch clamp recordings. I've used different sets of parameters to find the ones with the highest correlation between inferred and 'real' spike trains. I've found the best performance is with a tau of 0.8 and and estimated firing rate of 0.04. I've set the standard deviation of the measurement noise to the standard deviation over the lowest 50% of points of the dF/F trace.

If I then run fast-OOPSI on the dataset the correlation between the true and inferred spike train is very good (similar to Theis et al 2016) and there is a nice linear relationship between inferred spike train and actual number of spikes in a 100 ms timebin. I've included my findings as a figure, so you can check it out.

Thanks for making this code publicly available! [image: spikeinferenceoopsi-01] https://cloud.githubusercontent.com/assets/19360723/18956914/74d08adc-865d-11e6-8f17-b1069797faca.jpg

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/jovo/oopsi/issues/5#issuecomment-250473442, or mute the thread https://github.com/notifications/unsubscribe-auth/AACjch2YCMjwqnStZhu_ABcabI7108ygks5qu8PegaJpZM4KJ75q .

the glass is all full: half water, half air. neurodata.io, jovo calendar https://calendar.google.com/calendar/embed?src=joshuav%40gmail.com&ctz=America/New_York