MountainSort MDA sorting data conversion

Install NeuroConv with the additional dependencies necessary for reading MountainSort MDA sorting data.

pip install "neuroconv[mda]"

Convert MountainSort firings.mda sorting output to NWB using MdaSortingInterface.

The MDA sorting format (MountainSort v4 and earlier) stores, per spike event, the primary channel, sample time, and unit label. See the MountainSort firings.mda specification for details. This interface reads firings.mda files produced by ml_ms4alg (MountainSort v4 standalone) or MountainLab-js pipelines. SpikeInterface-wrapped MountainSort workflows produce firings.npz instead and are not served by this interface.

>>> from datetime import datetime
>>> from zoneinfo import ZoneInfo
>>>
>>> from neuroconv.datainterfaces import MdaSortingInterface
>>>
>>> interface = MdaSortingInterface(file_path=mda_firings_path, sampling_frequency=mda_sampling_frequency, verbose=False)
>>>
>>> metadata = interface.get_metadata()
>>> session_start_time = datetime(2020, 1, 1, 12, 30, 0, tzinfo=ZoneInfo("US/Pacific"))
>>> metadata["NWBFile"].update(session_start_time=session_start_time)
>>> # Add subject information (required for DANDI upload)
>>> metadata["Subject"] = dict(subject_id="subject1", species="Mus musculus", sex="M", age="P30D")
>>>
>>> nwbfile_path = f"{path_to_save_nwbfile}"  # This should be something like: "./saved_file.nwb"
>>> interface.run_conversion(nwbfile_path=nwbfile_path, metadata=metadata)

Linking units to their contact through peak channel

When the original recording is available, NWB allows each unit in the units table to be linked to the electrodes it is detected on through a DynamicTableRegion on units.electrodes (see How to Link Sorted Data to Electrodes). A common way to build this linkage is to assign each unit to the single electrode where its spikes are largest, that unit’s “peak channel.”

When MountainSort is run with peak-channel tracking enabled, firings.mda records this peak channel for each unit. However, the value is stored as a 1-indexed position into the subset of acquisition-system channels that were passed to the sorter, so it cannot be used directly to identify an electrode.

Neuroconv builds the linkage in terms of stable identifiers instead: the channel IDs of the recording and the unit IDs of the sorting, both provided through SpikeInterface. The missing piece is the bridge between the positional index in firings.mda and the recording’s channel IDs, namely the ordered list of channel IDs that was passed to the sorter when the algorithm was run.

When this information is available, SortedRecordingConverter can be used as follows:

from neuroconv.converters import SortedRecordingConverter
from neuroconv.datainterfaces import IntanRecordingInterface, MdaSortingInterface

recording_interface = IntanRecordingInterface(file_path="path/to/intan.rhd")
sorting_interface = MdaSortingInterface(
    file_path="path/to/firings.mda", sampling_frequency=30_000,
)

# The recording channel IDs corresponding to the channel indices that were
# passed to MountainSort, in the same order. When you ran MountainSort, you
# selected a subset of the recording's channels (e.g. indices [60, 61, 62, 63])
# and handed their data to the sorter; the IDs of those channels, in that order,
# go here. Parse from your pipeline config (e.g. .trodesconf for SpikeGadgets)
# or from the extraction script that built raw.mda.
sorter_input_channel_ids = ["A-060", "A-061", "A-062", "A-063"]

unit_ids = sorting_interface.sorting_extractor.get_unit_ids()
peak_channel_indices = sorting_interface.sorting_extractor.get_property("mda_peak_channel")

unit_ids_to_channel_ids = {}
for unit_id, peak_channel_index in zip(unit_ids, peak_channel_indices):
    # MountainSort stores peak channels as 1-indexed positions; subtract 1 for 0-indexed list access
    position_in_sorter_input = int(peak_channel_index) - 1
    peak_channel_id = sorter_input_channel_ids[position_in_sorter_input]
    unit_ids_to_channel_ids[unit_id] = [peak_channel_id]

converter = SortedRecordingConverter(
    recording_interface=recording_interface,
    sorting_interface=sorting_interface,
    unit_ids_to_channel_ids=unit_ids_to_channel_ids,
)
nwbfile = converter.create_nwbfile()

If either piece is missing, the peak-channel value remains on the units table as a plain integer column named mda_peak_channel and the formal electrode linkage is simply skipped.