Skip to main content

Sequence Detection (Offline Use)

RT-Sort can be used for offline sequence detection with a pre-trained detection model. Here's how to do it for both Maxwell MEAs and Neuropixels.

Maxwell MEAs​

from braindance.core.spikesorter.rt_sort import detect_sequences

rt_sort = detect_sequences(recording, inter_path, detection_model, **kwargs)

Parameters​

  • recording: Recording loaded with SpikeInterface or a path to a saved recording (str or pathlib.Path).
  • inter_path: Path to a folder where RT-Sort's intermediate cached data is stored.
  • detection_model: ModelSpikeSorter object or a path to a folder containing a ModelSpikeSorter object's files.

Optional Parameters​

  • recording_window_ms: Tuple (start_ms, end_ms) indicating which section of the recording to run RT-Sort. Default is None (entire duration).
  • delete_inter: Whether to delete the directory inter_path and its contents. Default is False.
  • verbose: Whether to print progress of RT-Sort. Default is True.
  • num_processes: Number of CPU processes to use. Default is None (uses all available logical CPUs).

Returns​

  • If return_spikes=False (default): Returns an RTSort object.
  • If return_spikes=True: Returns a NumPy array of shape (num_sequences,) where each element is a NumPy array containing a sequence's spike train.

Example​

# Detect sequences in the first 5 minutes of a recording
rt_sort = detect_sequences(recording, inter_path, detection_model, recording_window_ms=(0, 5*60*1000))

# Assign spikes in the next 5 minutes
sequence_spike_trains = rt_sort.sort_offline(recording, inter_path, recording_window_ms=(5*60*1000, 10*60*1000), verbose=True)

Neuropixels​

The process for Neuropixels is similar to Maxwell MEAs, with one key difference:

from braindance.core.spikesorter.rt_sort import detect_sequences, neuropixels_params

rt_sort = detect_sequences(recording, inter_path, detection_model, **neuropixels_params, **kwargs)

Use the neuropixels_params dictionary to pass Neuropixels-specific RT-Sort parameters. It's also recommended to use a Neuropixels-specific detection model.

Next Steps​

After detecting sequences, you can proceed to real-time application or training your own models.