Two-photon calcium imaging dataset
https://drive.google.com/drive/folders/1MVDSZs-6K-tGb79wU9wHuQ2Veoho-LiP?usp=sharing
Raw two-photon calcium images are processed using suite2p. It extracts fluorescence traces of all the segmented neurons.\
- https://suite2p.readthedocs.io/en/latest/outputs.html
- https://www.biorxiv.org/content/10.1101/061507v2.abstract
- ffneu_z_sel.npy
- fluoresence traces. (num_neuron, len_seq_neural)
- f-fneu*0.7. f:fluorescence traces, fneu:neuropil fluorescence traces.
- Preprocess: Z-scored (z) and neurons with high possibility are selected (sel).
- spks_z_sel.npy
- deconvolved traces. Spikes. (num_neuron, len_seq_neural)
- Deconvolution means what?: https://suite2p.readthedocs.io/en/latest/FAQ.html
- Preprocess: Z-scored (z) and neurons with high possibility are selected (sel).
- The stream of data processing can be represented as: Raw two-photon calcium images -> ffneu -> spks.
- In Dadarlatlab's preliminary research, only spks worked for decoding (mapping spks->limb coordinates). So I would recommend to start with spks first. It seems that ffneu is too noisy to relate to behavior, or we don't have enough big size of data.
- But our future plan includes to use raw calcium images for real-time application.
- Raw calcium images are too large so not included here. Ask Seungbin if you want files.
Limb coordinates were predicted from video frames using Deeplabcut.\
- behav_coord_likeli.npy
- Continuous limb coordinates of a running mouse. (8, len_seq_behav). 8 = x and y coordinate for each four limb (right front, right hind, left front, left hind)
- Preprocess: Coordinates with low likelihood from Deeplabcut (reference Deeplabcut pages) were replaced to interpolated values with neighboring coordinates (likeli).
- behav_coord_likeli_norm.npy
- Preprocess: Min-max scaled (norm).
- GNN for Classification
- Transformer for reconstruction and subsequent classification
- Encoder for limb ---> neuron
