README.md

metadata

license: cc-by-4.0
task_categories:
  - time-series-forecasting
  - tabular-regression
tags:
  - neuroscience
  - electrophysiology
  - striatum
  - Neuropixels
  - neural-decoding
  - MitoPark
  - mouse
  - locomotion
configs:
  - config_name: noBin
    data_files:
      - split: train
        path: noBin/train-*
  - config_name: 1ms
    data_files:
      - split: train
        path: data/1ms/*.parquet
  - config_name: 2ms
    data_files:
      - split: train
        path: data/2ms/*.parquet
  - config_name: 5ms
    data_files:
      - split: train
        path: data/5ms/*.parquet
  - config_name: 10ms
    data_files:
      - split: train
        path: data/10ms/*.parquet
  - config_name: 25ms
    data_files:
      - split: train
        path: data/25ms/*.parquet
  - config_name: 50ms
    data_files:
      - split: train
        path: data/50ms/*.parquet
  - config_name: 100ms
    data_files:
      - split: train
        path: data/100ms/*.parquet
  - config_name: 200ms
    data_files:
      - split: train
        path: data/200ms/*.parquet
  - config_name: 500ms
    data_files:
      - split: train
        path: data/500ms/*.parquet

MitoPark Striatal Neural-Locomotion Dataset

Overview

High-density extracellular recordings from mouse brain during treadmill locomotion, designed for neural decoding of movement kinematics.

Subsets

Subset	Description	Rows
noBin	Raw spikes (session, neuron_id, spike_time, speed)	130,321,346
1ms	Binned at 1ms	-
2ms	Binned at 2ms	-
5ms	Binned at 5ms	-
10ms	Binned at 10ms	-
25ms	Binned at 25ms	-
50ms	Binned at 50ms	-
100ms	Binned at 100ms	-
200ms	Binned at 200ms	-
500ms	Binned at 500ms	-

Data Acquisition

Parameter	Value
Recording system	Neuropixels 1.0
Target regions	DLS (dorsolateral striatum), DMS (dorsomedial striatum)
Animal model	MitoPark mouse (progressive dopaminergic neurodegeneration)
Sessions	73 recordings
Neurons	100-339 units per session

Processing Pipeline

Spike Sorting

• Software: Kilosort 4
• Curation: Automated quality metrics (ISI violations, SNR, amplitude stability)
• Output: pixelCluster_imec0.mat containing spike times, cluster IDs, waveform metrics, and channel depth

Behavioral Signal

• Source: Analog belt encoder (25 kHz native sampling)
• Calibration: V₀ baseline correction via adaptive rolling window (60s) from detected immobility periods
• Conversion: 40 cm/s/V (Teensy calibration)
• Filtering: 8th-order Butterworth low-pass (5 Hz cutoff)
• Smoothing: 100 ms moving average
• Final sampling: 200 Hz
• Output: belt_speed.mat containing aligned speed trace (cm/s)

Dataset Generation

Raw spikes (noBin)

Each spike event paired with interpolated speed at spike time.

Column	Type	Description
session	string	Recording session identifier
neuron_id	int32	Neuron index (1-indexed)
spike_time	float32	Spike timestamp (seconds)
speed	float32	Treadmill speed at spike time (cm/s)

Binned subsets (1ms to 500ms)

Spike counts aggregated in temporal bins with lag structure for temporal modeling.

Column	Type	Description
N{i}_t2	float32	Neuron i spike count, 2 bins prior
N{i}_t1	float32	Neuron i spike count, 1 bin prior
N{i}_t	float32	Neuron i spike count, current bin
speed	float32	Mean treadmill speed in current bin (cm/s)

Structure: [N1_t2, N1_t1, N1_t, N2_t2, N2_t1, N2_t, ..., Nn_t2, Nn_t1, Nn_t, speed]

Usage

from datasets import load_dataset

# Raw spike events
ds = load_dataset("opsecsystems/mitopark-striatal-neural", "noBin")

# Binned at 5ms resolution
ds = load_dataset("opsecsystems/mitopark-striatal-neural", "5ms")

# Access data
X = ds['train'].to_pandas().drop('speed', axis=1).values
y = ds['train']['speed']

File Structure

mitopark-striatal-neural/
├── noBin/
│   └── train-00000-of-00001.parquet
├── data/
│   ├── 1ms/*.parquet
│   ├── 2ms/*.parquet
│   ├── 5ms/*.parquet
│   ├── 10ms/*.parquet
│   ├── 25ms/*.parquet
│   ├── 50ms/*.parquet
│   ├── 100ms/*.parquet
│   ├── 200ms/*.parquet
│   └── 500ms/*.parquet
└── README.md

Citation

@dataset{mitopark_striatal_neural_2025,
  author = {Leutgeb Laboratory},
  title = {MitoPark Striatal Neural-Locomotion Dataset},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/opsecsystems/mitopark-striatal-neural}
}

License

CC-BY-4.0

Contact

Leutgeb Laboratory, UC San Diego