科研 Skill

细胞外记录分选质控

neuro-spike-sorting-qc

Guides spike sorting and quality control of sorted units from extracellular recordings with explicit, reproducible metrics. Use when curating clusters or tracking units across days.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

Spike sorting and quality control

Sorter output is a list of candidate clusters, not neurons. Decide which are well isolated single units with explicit metrics, applied the same way to every session, and report the thresholds.

0. Load the tool reference

TaskReference
Read, preprocess, sort, analyse, compare with ground truth in SpikeInterfacereferences/spikeinterface.md
Kilosort 4 settings, outputs, checksreferences/kilosort.md
Metrics, Bombcell, UnitRefine, merges, Phy, UnitMatchreferences/curation.md
Neuropixels: SpikeGLX, Open Ephys, phase shift, drift, CatGT/TPrime sync, probe localisationreferences/neuropixels.md

Each has a recipe checked against the current library, its traps, and what to report. Loading other acquisition formats: neuro-file-formats.

1. Before sorting

  • Check the probe geometry file matches the hardware (site positions, channel order; see neuro-channel-map-verifier).
  • Common-average or common-median reference and high-pass (about 300 Hz) as the sorter expects.
  • Estimate and correct drift (Kilosort and SpikeInterface motion correction); look at the drift map before trusting long recordings.

2. Compute standard metrics per unit

Use SpikeInterface compute_quality_metrics or Bombcell:

MetricMeasuresCommonly used threshold
ISI violations (refractory, about 1.5 ms)contamination by other spikesviolation ratio below 0.5, or contamination estimate below about 10%
Amplitude cutoffspikes lost below detection thresholdbelow 0.1
Presence ratiounit present across the sessionabove 0.9
Firing ratevery low rates make metrics unreliableabove about 0.1 Hz
SNR / amplitudeseparation from noisedataset dependent
Isolation distance, L-ratio, d'cluster separation in feature spacerelative, compare within dataset
Driftposition change over timeflag large drift

The thresholds in the table follow commonly used defaults (Allen Institute pipelines; Siegle et al. 2021) and are a starting point, not a rule. State the ones you use.

3. Curate

  1. Auto-label with metrics (good / MUA / noise).
  2. Inspect borderline units in Phy or SpikeInterface GUI: waveform shape and spatial footprint, autocorrelogram (refractory dip), amplitude over time (drift or loss), cross-correlograms with neighbours (merge candidates show a refractory dip in the cross-correlogram).
  3. Remove noise clusters (non-physiological waveforms, waveforms on all channels, line-frequency locked).
  4. Log every manual merge or split.

4. Report

Sorter and version, parameters, drift correction, metric thresholds, counts of good/MUA/noise units per session, and which analyses use only single units vs multi-unit activity.

References

Hill, Mehta & Kleinfeld 2011 (quality metrics); Siegle et al. 2021 (Allen Brain Observatory); Fabre et al. 2023 (Bombcell); Buccino et al. 2020 (SpikeInterface); Pachitariu et al. 2024 (Kilosort4).

示例提示词

帮我用Kilosort4对这份Neuropixels数据做分选并质控单元可信度。

所属插件

神经科学面向神经科学研究:EEG/MEG、fMRI、BIDS、spike sorting、连接组、脑机接口解码评估、神经刺激安全与公开神经数据检索。

安装该插件会一并获得同领域的 45 个 Skill / MCP。

同插件的其他科研 Skill

常见问题

细胞外记录分选质控能做什么?

Guides spike sorting and quality control of sorted units from extracellular recordings with explicit, reproducible metrics. Use when curating clusters or tracking units across days。

如何安装细胞外记录分选质控?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「细胞外记录分选质控」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

细胞外记录分选质控需要哪些权限?

该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。

细胞外记录分选质控经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。