README.md

metadata

configs:
  - config_name: worm_data_short
    data_files:
      - split: train
        path: worm_data_short.parquet
language:
  - en
license: mit

CITATION

Q. Simeon, L. Venâncio, M. A. Skuhersky, A. Nayebi, E. S. Boyden and G. R. Yang, "Scaling Properties for Artificial Neural Network Models of a Small Nervous System," SoutheastCon 2024, Atlanta, GA, USA, 2024, pp. 516-524, doi: 10.1109/SoutheastCon52093.2024.10500049.

DATASET PROCESSING

worm_data_short.parquet is generated by aggregating the info from 12 neural activity source datasets. Each source dataset is processed as follows:

1. Loading raw data in various formats (MATLAB files, JSON files, etc.).
2. Extracting relevant data fields (neuron IDs, traces, time vectors, etc.).
3. Cleaning data
4. Resampling the data to a common time resolution. - if requested
5. Smoothing the data using different methods - if requested
6. Normalizing data
7. Creating dictionaries to map neuron indices to neuron IDs and vice versa.
8. Saving the preprocessed data into a standardized format.

DATASET CONFIG

This dataset was preprocessed with the following hyperparameters. To modify or reproduce the dataset with new settings, refer to the source code.

• resample_dt: 0.333 — Time step for resampling
• interpolate: "linear" — Method used to fill missing data
• smooth:
  • method: "moving" — Smoothing algorithm (none, gaussian, exponential, moving)
  • alpha: 0.5 — Exponential smoothing factor
  • sigma: 5 — Gaussian kernel width
  • window_size: 15 — Window size for moving average
• norm_transform: "standard" — Type of normalization (standard or causal)

FIGURE

Compiled neural activity dataset from GCaMP calcium imaging of C. elegans from multiple experimental sources, standardized to a common sampling rate and organization format.

EXAMPLE USAGE

To use this dataset, you may load or download it using the datasets and huggingface_hub libraries, respectively.

See this notebook on an example of how to download the dataset and begin working with the data.

Google Colab notebook example of loading this dataset and then plotting a few samples of calcium data.

ORIGINAL DATA FILES

We provide a Dropbox link to download the original data that we obtained from various sources, including publicly available and unpublished data shared with us by researchers. The raw_data_file in the dataset table references these files.

If you'd like to preprocess the data from scratch using different preprocessing settings or datasets, you may do so using the code in the worm-data-preprocess repo.