Synthetic Tetrode Recording Dataset With Spike-Waveform Drift Over 256 Hours

A synthetic ground-truth dataset modeling long-term extracellular tetrode recordings from the rodent brain over 256 hours. It comprises spiking activity for 8 distinct single-units with drifting amplitudes, created by Ashesh K. Dhawale of Harvard University to test spike-sorting algorithms. The dataset includes spike times, waveform snippets, and simulation parameters.

Use Cases

• Benchmarking spike-sorting algorithm accuracy based on synthetic ground-truth spike times and unit identities.
• Testing clustering performance on non-stationary data based on drifting single-unit spike amplitudes.
• Evaluating algorithm robustness based on background multi-unit spiking activity at 20 Hz.
• Simulating long-term neural recordings based on a 256-hour continuous synthetic timescale.

Strengths

• Provides ground-truth labels for spike times and unit identities, enabling precise algorithm validation.
• Models realistic non-stationarity with single-unit spike amplitudes drifting over a range of 100 to 400 μV.
• Simulates a long, continuous recording period of 256 hours at a 30 kHz sampling rate.
• Includes parameters for a homogenous Poisson process with unit-specific firing rates and a 2 ms refractory period.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Row count is unknown, which may limit suitability assessment.
• The provided sample contains data from 1 tetrode, with the full 6-tetrode dataset requiring separate access.

Provenance

Source

Harvard University, Olveczky Lab

Collection Method

Synthetic generation modeling recordings from rodent motor cortex and striatum.

Time Range

Synthetic recording spans 256 hours.