Requested Data

[pauladkisson]

Important data for the datta-lab-to-nwb conversion that is not yet provided.

Fiber Photometry

• [x] Commanded Voltage Series from the TDT
• [x] Alignment slopes and/or code

Behavior

• [x] Raw Depth Video from Kinect

Optogenetics

• [x] Raw power series from Opto Engine MRL-III (esp. for 3s-pulsed stim)
  • Not Available

Velocity-Modulation

• [x] Position info for velocity-modulation experiments
• [ ] Raw data for velocity modulation is 90mins, but the processed data is only 45 mins long. Where does that fit in terms of temporal alignment?

Keypoints

• [x] Raw data for keypoint experiments
• [ ] Session start times for keypoint experiments
• [ ] Which videos correspond to which column in matched_timestamps
• [ ] Reference frame
• [ ] alignment_df.parquet file with slope and bias terms for temporal alignment

[kaijfox]

Sorry for the delay here but we've tracked down most of this now. A couple of quick things:

• Could you clarify what you're looking for in terms of reference frames? I suspect it'll be there in the data we've got lined up to pass you but I want to make sure and be able to point you to the right file.
• Unfortunately we didn't record raw opto power series.
• I believe the position info for the velocity experiments is supposed to be in the dataframes on already zenodo. Is it missing? Or do we need something more specific than what's there?
• Do you have a standard way you like to receive raw data like this? We're looking at ~11TB.

Thanks!

[pauladkisson]

Hey @kaijfox, good to hear from you!

Could you clarify what you're looking for in terms of reference frames?

The reference_frame is a description of the "zero-position" for a Spatial Series like the mouse's position or orientation. For example, "top left", "bottom right", etc.

Unfortunately we didn't record raw opto power series.

No worries, we will just have to work with the data we have.

I believe the position info for the velocity experiments is supposed to be in the dataframes on already zenodo.

XY-position information (centroid_x_mm, centroid_y_mm) is not present in the optoda_raw_data/closed_loop_behavior_velocity_conditioned.parquet dataframe and I didn't see any other files that related to the velocity modulation experiments. If I am overlooking a file already in the zenodo please lmk.

Do you have a standard way you like to receive raw data like this? We're looking at ~11TB.

I think this google drive folder should be able to handle all the data, but I will double-check with @bendichter.

If you get a chance, please also take a look at some of the clarifying questions (#14), which will help me a lot to move forward on this conversion!

Thanks!

[CodyCBakerPhD]

I think this google drive folder should be able to handle all the data, but I will double-check with @bendichter.

The drive is where it will end up, yes - but @bendichter will need to set up a Globus transfer link for them to that location, otherwise if you try to 'drag-and-drop' or other drive-specific methods the storage will count as personal usage (capped at whatever plan the user has) instead of our usage (uncapped)

It's also bandwidth throttled, so you'll only be able to transfer about 1 TB a day - but as long as both Globus endpoints remain active, you should only need to trigger the transfer once and it will be smart enough to reattempt every day

[kaijfox]

Great a globus endpoint will be very helpful. Once we get that I'll upload a couple sessions from each experiment so you have some stuff to work with right away before we hit the bandwidth limit and then we can fill it out with the rest of the data as a second step.

On the reference_frame issue, I'm going to include that in #14, which Win is going to look at in the next couple days since he was around for the data collection.

I'll look into the position information for closed_loop_behavior_velocity_conditioned.parquet. It seems likely that the info simply wasn't included there because it's not directly used in the analysis, but it will be present in moseq-extract data once that's on the google drive so I'll let you know which files to look in specifically.

[bendichter]

@kaijfox can you see if you have access to: https://app.globus.org/file-manager?origin_id=a0e2af7a-11c6-4118-b15a-f53c219e05e1&origin_path=%2FMy%20Drive%2Fdata%2FDatta-lab-to-nwb%2F

I just gave access to "Kai Jordan Fox (kfox@access-ci.org)"

[kaijfox]

Ah I believe I'm kfox1@access-ci.org

[bendichter]

@kaijfox ok, added!

[kaijfox]

Preliminary raw data should be in that google drive.

The root directory contains a number of experiment directories, each of which contains several session directories. The experiments used in the analyses for the paper are listed in a TOML file analysis_configuration.TEMPLATE found in the root directory of dattalab/dopamine-reinforces-spontaneous-behavior.

Each session directory contains:

• proc, directory – Output of moseq-extract.
• depth.avi – Raw, lossless-compressed depth frames.
• depth_ts.txt – Timestamps for depth frames.
• feedback_ts.txt – Timestamps at which the target syllable was detected, and a boolean denoting whether or not reinforcement was given.
• A number of other intermediate processing files.

If the experiment included photometry, the session folder will contain:

• ir.avi – Infrared video containing LED flashes used to align video stream with photometry data.
• ir_ts.avi – Timestamps for frames in ir.avi.
• tdt_data_*.dat – Raw recording from the TDT.
• tdt_data_*.json - Metadata from the TDT recording.

A couple of notes

• All experiment directories are uploaded, but each contains only three sessions out of many that will be uploaded soon.
• We are still working on a mapping between the experiment groups in analysis_configuration.TEMPLATE and the parquet files on zenodo.
• The alignment of photometry data to depth/behavioral data occurs via IR leds visible in ir.avi. @bendichter noted you might be interested in what that process looked like as you're building out some general stream syncing functionality at the moment. Here are a couple of files from Win's analysis code (attached: sync_code.zip) and his description of the procedure:

  I have two IR LEDs inside the behavior bucket. The depth camera picks up the LED flashes, and the voltages supplied to the LEDs are copied to the TDT system so that they're aligned with the photometry signal.Here are two files from my codebase that I used to align the photometry signal to behavior. It starts with the function align_photometry_to_behavior_v2 in the align.py file. It then reads in the IR video, and automatically finds the pixels most likely corresponding to the IR leds in the detect_sync_led_from_behavior function in signal.py. Using those pixels, does some de-noising because there's a ton of noise in this signal, and computes the onsets/offsets of the LED flashes (throwing out any onsets/offsets that are too fast or slow). One of my LEDs is flashing at a constant 2s-on and 2s-off cycle, while the other one is flashing at a 1s-on and uniform random 1-~5s off cycle, so I try to figure out which LED is likeliest to be the random and uniform flasher. After that, I compare the onsets/offsets with the signal copied to the TDT. Since the detected onsets/offsets are so noisy, I do a refinement procedure to remove the least likely datapoints when fitting a linear map to go from behavior timestamps to photometry timestamps and vice versa.

I'm sure questions will come up as you dig into the files just let me know!

[bendichter]

@CodyCBakerPhD , @pauladkisson this looks like it might be a good opportunity for us to develop hooks in NeuroConv for custom temporal alignment procedures.