Curation Tutorial

After spike sorting and computing quality metrics, you can automatically curate the spike sorting output using the quality metrics that you have calculated.

Import the modules and/or functions necessary from spikeinterface

import spikeinterface.core as si

from spikeinterface.qualitymetrics import compute_quality_metrics

Let’s generate a simulated dataset, and imagine that the ground-truth sorting is in fact the output of a sorter.

recording, sorting = si.generate_ground_truth_recording()
print(recording)
print(sorting)

GroundTruthRecording (InjectTemplatesRecording): 4 channels - 25.0kHz - 1 segments
                      250,000 samples - 10.00s - float32 dtype - 3.81 MiB
GroundTruthSorting (NumpySorting): 10 units - 1 segments - 25.0kHz

Create SortingAnalyzer

For this example, we will need a SortingAnalyzer and some extensions to be computed first

analyzer = si.create_sorting_analyzer(sorting=sorting, recording=recording, format="memory")
analyzer.compute(["random_spikes", "waveforms", "templates", "noise_levels"])

analyzer.compute("principal_components", n_components=3, mode="by_channel_local")
print(analyzer)

estimate_sparsity (no parallelization):   0%|          | 0/10 [00:00<?, ?it/s]
estimate_sparsity (no parallelization): 100%|██████████| 10/10 [00:00<00:00, 417.36it/s]

compute_waveforms (no parallelization):   0%|          | 0/10 [00:00<?, ?it/s]
compute_waveforms (no parallelization): 100%|██████████| 10/10 [00:00<00:00, 324.70it/s]

noise_level (no parallelization):   0%|          | 0/20 [00:00<?, ?it/s]
noise_level (no parallelization): 100%|██████████| 20/20 [00:00<00:00, 263.04it/s]

Fitting PCA:   0%|          | 0/10 [00:00<?, ?it/s]
Fitting PCA: 100%|██████████| 10/10 [00:00<00:00, 180.81it/s]

Projecting waveforms:   0%|          | 0/10 [00:00<?, ?it/s]
Projecting waveforms: 100%|██████████| 10/10 [00:00<00:00, 2619.64it/s]
SortingAnalyzer: 4 channels - 10 units - 1 segments - memory - sparse - has recording
Loaded 5 extensions: random_spikes, waveforms, templates, noise_levels, principal_components

Then we compute some quality metrics:

metrics = compute_quality_metrics(analyzer, metric_names=["snr", "isi_violation", "nearest_neighbor"])
print(metrics)

calculate pc_metrics:   0%|          | 0/10 [00:00<?, ?it/s]
calculate pc_metrics:  60%|██████    | 6/10 [00:00<00:00, 51.29it/s]
calculate pc_metrics: 100%|██████████| 10/10 [00:00<00:00, 51.43it/s]
         snr  isi_violations_ratio  ...  nn_hit_rate  nn_miss_rate
0   2.081982                   0.0  ...     0.495536      0.068199
1   1.897748                   0.0  ...     0.496711      0.067587
2   0.603453                   0.0  ...     0.513793      0.039226
3  27.721287                   0.0  ...     0.975904      0.004222
4   2.954370                   0.0  ...     0.493421      0.054506
5  14.908217                   0.0  ...     0.916176      0.008468
6   9.532658                   0.0  ...     0.924837      0.005636
7   3.572693                   0.0  ...     0.743056      0.028360
8   1.941346                   0.0  ...     0.365809      0.061243
9  30.585269                   0.0  ...     0.963028      0.002165

[10 rows x 5 columns]

We can now threshold each quality metric and select units based on some rules.

The easiest and most intuitive way is to use boolean masking with a dataframe.

Then create a list of unit ids that we want to keep

keep_mask = (metrics["snr"] > 7.5) & (metrics["isi_violations_ratio"] < 0.2) & (metrics["nn_hit_rate"] > 0.90)
print(keep_mask)

keep_unit_ids = keep_mask[keep_mask].index.values
keep_unit_ids = [unit_id for unit_id in keep_unit_ids]
print(keep_unit_ids)

0    False
1    False
2    False
3     True
4    False
5     True
6     True
7    False
8    False
9     True
dtype: bool
['3', '5', '6', '9']

And now let’s create a sorting that contains only curated units and save it.

curated_sorting = sorting.select_units(keep_unit_ids)
print(curated_sorting)


curated_sorting.save(folder="curated_sorting")

GroundTruthSorting (UnitsSelectionSorting): 4 units - 1 segments - 25.0kHz

NumpyFolder (NumpyFolderSorting): 4 units - 1 segments - 25.0kHz

Unit IDs

['3' '5' '6' '9']

Annotations

• name : GroundTruthSorting

Properties

gt_unit_locations
[[12.629008 -7.895358 7.6198726] [22.504269 29.685955 18.601088 ] [29.937254 29.228424 43.268787 ] [ 7.276633 -9.420886 33.948456 ]]

We can also save the analyzer with only theses units

clean_analyzer = analyzer.select_units(unit_ids=keep_unit_ids, format="zarr", folder="clean_analyzer")

print(clean_analyzer)

SortingAnalyzer: 4 channels - 4 units - 1 segments - zarr - sparse - has recording
Loaded 6 extensions: random_spikes, waveforms, templates, noise_levels, principal_components, quality_metrics