Neuropixels 数据分析指南基于 SpikeInterface 的 SortingAnalyzer 后处理、质量指标与可视化全流程【免费下载链接】scientific-agent-skillsTurn any AI agent into an AI Scientist. The #1 Agent Skills library for science, used by 190,000 scientists worldwide. 165 ready-to-use validated skills plus 100 scientific databases covering biology, chemistry, medicine, and drug discovery. Compatible with Cursor, Claude Code, Codex, Pi, Antigravity, and the open Agent Skills standard.项目地址: https://gitcode.com/GitHub_Trending/cl/scientific-agent-skills导读本指南以 scientific-agent-skills 仓库中 skills/neuropixels-analysis 技能的 ANALYSIS.md 参考文档为核心系统讲解基于 SpikeInterface 完成 Neuropixels 高分拣结果后处理与分析的完整方案从SortingAnalyzer的创建与扩展计算、质量指标的定义与阈值筛选、波形/模板/单位位置/相关图分析到探针图、漂移图、单元摘要等可视化以及 LFP 分析、Phy/NWB/报告导出和端到端管线。读完本文你将能够独立搭建一条分拣 → 后处理 → 质量评估 → 可视化 → 导出的 Neuropixels 数据发布级分析流程并能直接使用本仓库配套的脚本与模板。从分拣结果到分析对象SortingAnalyzerSpike sorting 输出的sorting对象只包含哪个单元在哪个采样点放电这一最基本的时序信息。所有后续分析——波形、模板、噪声、PCA、相关图、质量指标——都需要把分拣结果与原始recording对齐计算这正是SortingAnalyzer的职责。它是 SpikeInterface 后处理的核心对象仓库中的 scripts/neuropixels_pipeline.py 与 assets/analysis_template.py 均围绕它组织整个后处理阶段。创建 Analyzercreate_sorting_analyzer接收分拣结果与预处理后的记录对象并以稀疏sparse方式存储只保留每个单元周围有信号的通道而不是整个探针的全部通道从而大幅压缩存储与计算量import spikeinterface.full as si # Create analyzer analyzer si.create_sorting_analyzer( sorting, recording, sparseTrue, # Use sparse representation formatbinary_folder, # Storage format folderanalyzer_output # Save location )关键参数说明sparseTrue按单元稀疏存储波形Neuropixels 1.0/2.0 拥有 384 个记录通道若全部密排存储会非常浪费format支持memory、binary_folder与zarr持久化分析建议使用binary_folder磁盘目录需要流式/大数组场景可改用 zarrfolder分析结果的落盘位置。注意run_sorter与create_sorting_analyzer使用folder参数而导出函数使用output_folder见 references/api_reference.md。计算扩展ExtensionsSortingAnalyzer的一切分析能力都通过扩展extension提供每次compute会生成一类可独立存取的分析数据# Compute all standard extensions analyzer.compute(random_spikes) # Random spike selection analyzer.compute(waveforms) # Extract waveforms analyzer.compute(templates) # Compute templates analyzer.compute(noise_levels) # Noise estimation analyzer.compute(principal_components) # PCA analyzer.compute(spike_amplitudes) # Amplitude per spike analyzer.compute(correlograms) # Auto/cross correlograms analyzer.compute(unit_locations) # Unit locations analyzer.compute(spike_locations) # Per-spike locations analyzer.compute(template_similarity) # Template similarity matrix analyzer.compute(quality_metrics) # Quality metrics # Or compute multiple at once analyzer.compute([ random_spikes, waveforms, templates, noise_levels, principal_components, spike_amplitudes, correlograms, unit_locations, quality_metrics ])扩展之间存在依赖顺序random_spikes必须先于waveforms波形从随机抽取的尖峰中截取waveforms又先于templates模板是波形的聚合统计而principal_components先于依赖 PCA 的质量指标isolation_distance、l_ratio、d_prime、nn_hit_rate。仓库脚本 scripts/neuropixels_pipeline.py 正是按此顺序分组计算的analyzer.compute(random_spikes, methoduniform, max_spikes_per_unit500) analyzer.compute(waveforms, ms_before1.5, ms_after2.0, **job_kwargs) analyzer.compute(templates, operators[average, std]) analyzer.compute(noise_levels) analyzer.compute(spike_amplitudes, **job_kwargs) analyzer.compute(correlograms, window_ms100, bin_ms1) analyzer.compute(unit_locations, methodmonopolar_triangulation) analyzer.compute(template_similarity)保存与加载# Save analyzer.save_as(folderanalyzer_saved, formatbinary_folder) # Load analyzer si.load_sorting_analyzer(analyzer_saved)已计算的扩展会被持久化加载后无需重新计算即可直接读取这是避免对几十 GB 原始数据反复重算的关键。此外analyzer.select_units(unit_ids, folder..., formatbinary_folder)可用于在筛选后生成只含合格单元的干净 analyzer对应 SKILL.md 中的analyzer_clean导出流程。质量指标量化单元纯度、完整性与稳定性质量指标是分拣结果能否进入下游分析乃至论文的通行证。仓库 references/QUALITY_METRICS.md 将指标归纳为三类问题类别回答的问题关键指标污染Type I是否混入多个神经元ISI violations、SNR完整性Type II是否漏掉了尖峰Amplitude cutoff、presence ratio稳定性单元是否随时间漂移Drift 指标、amplitude CV计算指标analyzer.compute(quality_metrics) qm analyzer.get_extension(quality_metrics).get_data() print(qm)返回的qm是 pandas DataFrame每行对应一个单元、每列对应一项指标可直接进行query筛选或.to_csv()落盘。也可以只算需要的子集analyzer.compute( quality_metrics, metric_names[snr, isi_violations_ratio, presence_ratio, firing_rate] )完整指标速查表MetricDescriptionGood ValuessnrSignal-to-noise ratio 5isi_violations_ratioISI violation ratio 0.01 (1%)isi_violations_countISI violation countLowpresence_ratioFraction of recording with spikes 0.9firing_rateSpikes per second0.1-50 Hzamplitude_cutoffEstimated missed spikes 0.1amplitude_medianMedian spike amplitude-amplitude_cvCoefficient of variation 0.5drift_ptpPeak-to-peak drift (um) 40drift_stdStandard deviation of drift 10drift_madMedian absolute deviation 10sliding_rp_violationSliding refractory period 0.05sync_spike_2Synchrony with other units 0.5isolation_distanceMahalanobis distance 20l_ratioL-ratio (isolation) 0.1d_primeDiscriminability 5nn_hit_rateNearest neighbor hit rate 0.9nn_miss_rateNearest neighbor miss rate 0.1silhouette_scoreCluster silhouette 0.5对部分指标的阈值解读可参考 references/QUALITY_METRICS.md 的分档例如isi_violations_ratio按 0.01优良单单元、0.01–0.1轻微污染、0.1–0.5疑似多单元、 0.5极可能多单元分级其生理依据是神经元约 1.5 ms 的不应期Hill et al. 2011snr按 10优秀、5–10良好、2–5可接受、 2可能是噪声分级。参数化指标计算部分指标支持自定义参数例如不应期阈值、PCA 近邻数、峰值极性等# 自定义不应期 analyzer.compute(quality_metrics, metric_names[isi_violations_ratio], isi_threshold_ms1.5, min_isi_ms0.0) # PCA 空间隔离指标 analyzer.compute(quality_metrics, metric_names[isolation_distance], n_neighbors4) # 幅度截断的峰值极性 analyzer.compute(quality_metrics, metric_names[amplitude_cutoff], peak_signneg) # neg, pos, or both # 存在率的时间窗 analyzer.compute(quality_metrics, metric_names[presence_ratio], bin_duration_s60) # 1-minute bins # 漂移指标 analyzer.compute(quality_metrics, metric_names[drift_ptp, drift_std, drift_mad])依赖 PCA 的隔离类指标isolation_distance、l_ratio、d_prime、nn_hit_rate、silhouette_score需先计算principal_components默认n_components5。一次性计算全部指标时可按 references/QUALITY_METRICS.md 中的all_metric_names列表组织放电属性firing_rate、presence_ratio、波形snr、amplitude_cutoff、amplitude_cv_median、amplitude_cv_range、ISIisi_violations_ratio、isi_violations_count、漂移drift_ptp、drift_std、drift_mad、隔离PCA 类、同步sync_spike_2/4/8。自定义质量阈值筛选拿到指标表后即可定义自己的质量标准并筛选合格单元qm analyzer.get_extension(quality_metrics).get_data() # Define quality criteria quality_criteria { snr: (, 5), isi_violations_ratio: (, 0.01), presence_ratio: (, 0.9), firing_rate: (, 0.1), amplitude_cutoff: (, 0.1), } # Filter good units good_units qm.query( (snr 5) (isi_violations_ratio 0.01) (presence_ratio 0.9) ).index.tolist() print(fGood units: {len(good_units)}/{len(qm)})业界常用三套标准详见 references/QUALITY_METRICS.mdAllen Institute宽松presence_ratio 0.95 and isi_violations_ratio 0.5 and amplitude_cutoff 0.1IBL严格presence_ratio 0.9 and isi_violations_ratio 0.1 and amplitude_cutoff 0.1 and firing_rate 0.1Strict 单单元用于精确时序分析snr 5 and presence_ratio 0.99 and isi_violations_ratio 0.01 and amplitude_cutoff 0.01 and isolation_distance 20 and drift_ptp 40。仓库 scripts/compute_metrics.py 将这三套标准固化为CURATION_CRITERIA预设注意该脚本中snr依次为 3.0 / 4.0 / 5.0isi_violations_ratio依次为 0.5 / 0.1 / 0.01并对每个单元输出good/mua/noise标签tests/neuropixels-analysis/test_scripts.py 中的测试用例验证了能通过所有预设的单元必须是一个真正优秀单元EXEMPLARY以及strict 至少与 ibl 一样严格、ibl 至少与 allen 一样严格这一单调性约束防止阈值漂移造成静默的科学错误。波形与模板分析波形是单元信号的基本特征模板则是波形的统计聚合用于比较、分类与可视化。提取波形analyzer.compute(waveforms, ms_before1.5, ms_after2.5, max_spikes_per_unit500) # Get waveforms for a unit waveforms analyzer.get_extension(waveforms).get_waveforms(unit_id0) print(fShape: {waveforms.shape}) # (n_spikes, n_samples, n_channels)ms_before/ms_after决定截取窗口max_spikes_per_unit限制每个单元抽取的尖峰数默认约 500 足以稳定估计波形同时控制计算量。返回数组形状为(n_spikes, n_samples, n_channels)。计算模板analyzer.compute(templates, operators[average, std, median]) # Get template templates_ext analyzer.get_extension(templates) template templates_ext.get_unit_template(unit_id0, operatoraverage)operators可同时计算average、std、median等多种聚合std模板可直接用于绘制均值 ± 标准差的误差带references/plotting_guide.md 中的多面板单元摘要即用get_unit_template(unit_id, operatorstd)填充置信区间。模板相似度analyzer.compute(template_similarity) sim analyzer.get_extension(template_similarity).get_data() # Matrix of cosine similarities between templates该扩展输出单元模板两两之间的余弦相似度矩阵用于识别可能重复分拣的单元对或高度相似的模板。单元位置分析计算位置analyzer.compute(unit_locations, methodmonopolar_triangulation) locations analyzer.get_extension(unit_locations).get_data() print(locations) # x, y coordinates per unit尖峰位置analyzer.compute(spike_locations, methodcenter_of_mass) spike_locs analyzer.get_extension(spike_locations).get_data()位置方法选择center_of_mass快速、精度较低monopolar_triangulation更精确、较慢grid_convolution精度与速度的平衡选择。unit_locations给出每个单元在探针平面上的 x、y 坐标可叠加在探针图上查看单元的空间分布spike_locations则给出单个尖峰的位置可用于绘制尖峰位置散布图见 references/plotting_guide.md 的plot_spike_locations用法。相关图分析analyzer.compute(correlograms, window_ms50, bin_ms1) correlograms, bins analyzer.get_extension(correlograms).get_data() # correlograms shape: (n_units, n_units, n_bins) # Auto-correlogram for unit i: correlograms[i, i, :] # Cross-correlogram units i,j: correlograms[i, j, :]相关图是三维张量(n_units, n_units, n_bins)。对角切片[i, i, :]是单元 i 的自相关图autocorrelogram用于检查不应期凹槽refractory dip判断单元是否干净非对角切片[i, j, :]是单元对互相关图cross-correlogram可揭示单元间的功能连接或共享噪声源。window_ms为分析窗口宽度bin_ms为直方图分辨率。可视化从探针图到单元摘要SpikeInterface 的绘图 API 统一挂载在si命名空间下底层为spikeinterface.widgets as sw见 references/api_reference.md一行调用即可产出出版级图表。仓库 references/plotting_guide.md 还提供了完整的 Matplotlib 出版配置dpi、字体、单栏/双栏尺寸、色盲友好配色。探针与单元模板si.plot_probe_map(recording, with_channel_idsTrue) # All units si.plot_unit_templates(analyzer) # Specific units si.plot_unit_templates(analyzer, unit_ids[0, 1, 2])波形# Plot waveforms with template si.plot_unit_waveforms(analyzer, unit_ids[0]) # Waveform density si.plot_unit_waveforms_density_map(analyzer, unit_id0)Raster 图与幅度si.plot_rasters(sorting, time_range(0, 10)) # First 10 seconds analyzer.compute(spike_amplitudes) si.plot_amplitudes(analyzer) # Distribution si.plot_all_amplitudes_distributions(analyzer)相关图# Auto-correlograms si.plot_autocorrelograms(analyzer, unit_ids[0, 1, 2]) # Cross-correlograms si.plot_crosscorrelograms(analyzer, unit_ids[0, 1])质量指标可视化# Summary plot si.plot_quality_metrics(analyzer) # Specific metric distribution import matplotlib.pyplot as plt qm analyzer.get_extension(quality_metrics).get_data() plt.hist(qm[snr], bins50) plt.xlabel(SNR) plt.ylabel(Count)references/QUALITY_METRICS.md 还给出了多指标直方图矩阵2×3 子图、叠加阈值虚线以及发放率 vs SNR散点以 ISI 违例率为颜色映射等进阶方案用于批量审视所有单元的质量分布。位置、漂移与单元摘要si.plot_unit_locations(analyzer) si.plot_drift_raster(sorting, recording) # Comprehensive unit summary si.plot_unit_summary(analyzer, unit_id0)plot_unit_summary一张图集成波形、模板、自相关图、幅度、位置等关键信息是 AI 辅助审核与人工复核单元的最快入口仓库 references/AI_CURATION.md 描述的渲染单元摘要图 → 交由 Agent 评估隔离质量模式正依赖它。LFP 分析除尖峰外Neuropixels 的.lf.bin流低频带还提供局部场电位LFP可用于振荡、频谱与状态分析。加载 LFPlfp si.read_spikeglx(/path/to/data, stream_nameimec0.lf) print(fLFP: {lfp.get_sampling_frequency()} Hz)Neuropixels 1.0 的 AP 流采样率为 30 kHzLF 流为 2.5 kHz二者由同一.meta文件描述通过stream_name选择si.get_neo_streams(spikeglx, path)可先列出imec0.ap/imec0.lf/nidq等流见 references/api_reference.md。基础处理# Downsample if needed lfp_ds si.resample(lfp, resample_rate1000) # Common average reference lfp_car si.common_reference(lfp_ds, referenceglobal, operatormedian)降采样降低后续频谱计算成本全局中位参考可去除共同噪声源。提取波形与频谱分析import numpy as np # Get traces (channels x samples) traces lfp.get_traces(start_frame0, end_frame30000) # Specific channels traces lfp.get_traces(channel_ids[0, 1, 2])from scipy import signal import matplotlib.pyplot as plt # Get single channel trace lfp.get_traces(channel_ids[0]).flatten() fs lfp.get_sampling_frequency() # Power spectrum freqs, psd signal.welch(trace, fs, nperseg4096) plt.semilogy(freqs, psd) plt.xlabel(Frequency (Hz)) plt.ylabel(Power) plt.xlim(0, 100)频谱图f, t, Sxx signal.spectrogram(trace, fs, nperseg2048, noverlap1024) plt.pcolormesh(t, f, 10*np.log10(Sxx), shadinggouraud) plt.ylabel(Frequency (Hz)) plt.xlabel(Time (s)) plt.ylim(0, 100) plt.colorbar(labelPower (dB))结果导出Phy、NWB 与报告导出到 Phy人工审核si.export_to_phy( analyzer, output_folderphy_export, compute_pc_featuresTrue, compute_amplitudesTrue, copy_binaryTrue ) # Then: phy template-gui phy_export/params.pyPhy 是电生理社区标准的人工审核工具compute_pc_featuresTrue与compute_amplitudesTrue提供特征空间供人工聚类修正。注意export_to_phy使用output_folder参数与run_sorter的folder不同参见 references/api_reference.md 的特别说明。导出到 NWB数据共享from spikeinterface.exporters import export_to_nwb export_to_nwb( recording, sorting, output.nwb, metadatadict( session_descriptionNeuropixels recording, experimenterName, labLab name, institutionInstitution ) )NWB 是神经数据标准格式附带的 metadata 会写入文件头部保证实验溯源信息完整。导出报告si.export_report( analyzer, output_folderreport, remove_if_existsTrue, formathtml )format支持html与png仓库 scripts/neuropixels_pipeline.py 使用formatpng生成可嵌入论文的图集。完整分析管线一个可复用的函数将以上所有环节串起来即构成一条发布级管线。references/ANALYSIS.md 提供了完整实现import spikeinterface.full as si def analyze_sorting(recording, sorting, output_dir): Complete post-processing pipeline. # Create analyzer analyzer si.create_sorting_analyzer( sorting, recording, sparseTrue, folderf{output_dir}/analyzer ) # Compute all extensions print(Computing extensions...) analyzer.compute([random_spikes, waveforms, templates, noise_levels]) analyzer.compute([principal_components, spike_amplitudes]) analyzer.compute([correlograms, unit_locations, template_similarity]) analyzer.compute(quality_metrics) # Get quality metrics qm analyzer.get_extension(quality_metrics).get_data() # Filter good units good_units qm.query( (snr 5) (isi_violations_ratio 0.01) (presence_ratio 0.9) ).index.tolist() print(fQuality filtering: {len(good_units)}/{len(qm)} units passed) # Export si.export_to_phy(analyzer, f{output_dir}/phy) si.export_report(analyzer, f{output_dir}/report) # Save metrics qm.to_csv(f{output_dir}/quality_metrics.csv) return analyzer, qm, good_units # Usage analyzer, qm, good_units analyze_sorting(recording, sorting, output/)仓库现成脚本无需从零搭建本仓库的 scripts 目录提供了完整工具链explore_recording.py快速检查记录流、通道、时长、坏道preprocess_recording.py自动化预处理run_sorting.py运行分拣器compute_metrics.py计算质量指标并按allen/ibl/strict预设筛选对应 SKILL.md 的 curation 步骤export_to_phy.py导出 Phy 供人工审核neuropixels_pipeline.py端到端一键管线python scripts/neuropixels_pipeline.py /path/to/spikeglx/data output/ --sorter kilosort4 --curation allenscripts/neuropixels_pipeline.py 的run_pipeline会依次执行数据加载自动识别 SpikeGLX/Open Ephys→ 预处理高通滤波、坏道去除、相位校正、公共中位参考→ 漂移检测峰检测 质心定位drift_estimate超过 20 μm 才触发运动校正→ 分拣 → 后处理 → 质量指标 → 阈值 curation → 导出Phy、PNG 报告、quality_metrics.csv、unit_labels.json、summary.json。其命令行支持--sorterkilosort4/kilosort3/spykingcircus2/mountainsort5、--stream、--no-motion-correction、--curation等选项。若需要完全可编辑的脚本assets/analysis_template.py 将全部参数集中在文件顶部的PARAMETERS区DATA_PATH、DATA_FORMAT、FREQ_MIN/FREQ_MAX、APPLY_PHASE_SHIFT、MOTION_PRESET、SORTER_PARAMS、CURATION_METHOD、N_JOBS等复制后改参数即可运行cp assets/analysis_template.py my_analysis.py # Edit the PARAMETERS section, then run python my_analysis.py最佳实践要点结合 SKILL.md 的常见陷阱清单实战中应注意分拣前务必检查漂移——漂移超过约 10 μm 会显著降低质量先detect_peakslocalize_peaks绘制漂移光栅图需要时再用correct_motion(rec, presetnonrigid_fast_and_accurate, foldermotion/)校正Neuropixels 1.0 必须做phase_shift修正 ADC 采样偏移保存预处理后的记录rec.save(folder..., formatbinary)避免重复计算Kilosort 也需要二进制文件Kilosort4 建议使用 GPUCPU 场景可选 SpykingCircus2、Mountainsort5、Tridesclous2自动化/模型化筛选只是起点边界单元应结合人工或 AI 复核仓库提供 references/AUTOMATED_CURATION.md 与 references/AI_CURATION.md 两套进阶方案记录阈值与模型 repo ID保证结果可复现关键实验导出到 Phy 保留人工监督。依赖安装Python ≥ 3.10推荐 uv参考 SKILL.md 的 Installation 章节核心为uv pip install spikeinterface[full] probeinterface neo分拣器按需追加kilosort、spykingcircus、mountainsort5模型化筛选需huggingface_hub与skopsAI 审核可选anthropic。正式管线建议锁定版本文档示例spikeinterface0.104.3、kilosort4.1.7、probeinterface0.3.2、neo0.14.4。深入阅读完整工作流references/standard_workflow.mdAPI 速查references/api_reference.md出版级绘图配方references/plotting_guide.md预处理细节references/PREPROCESSING.md分拣细节references/SPIKE_SORTING.md运动校正references/MOTION_CORRECTION.md质量指标深入references/QUALITY_METRICS.md自动化与模型化筛选references/AUTOMATED_CURATION.mdAI 辅助审核references/AI_CURATION.md【免费下载链接】scientific-agent-skillsTurn any AI agent into an AI Scientist. The #1 Agent Skills library for science, used by 190,000 scientists worldwide. 165 ready-to-use validated skills plus 100 scientific databases covering biology, chemistry, medicine, and drug discovery. Compatible with Cursor, Claude Code, Codex, Pi, Antigravity, and the open Agent Skills standard.项目地址: https://gitcode.com/GitHub_Trending/cl/scientific-agent-skills创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考
Neuropixels 数据分析指南基于 SpikeInterface 的 SortingAnalyzer 后处理、质量指标与可视化全流程【免费下载链接】scientific-agent-skillsTurn any AI agent into an AI Scientist. The #1 Agent Skills library for science, used by 190,000 scientists worldwide. 165 ready-to-use validated skills plus 100 scientific databases covering biology, chemistry, medicine, and drug discovery. Compatible with Cursor, Claude Code, Codex, Pi, Antigravity, and the open Agent Skills standard.项目地址: https://gitcode.com/GitHub_Trending/cl/scientific-agent-skills导读本指南以 scientific-agent-skills 仓库中 skills/neuropixels-analysis 技能的 ANALYSIS.md 参考文档为核心系统讲解基于 SpikeInterface 完成 Neuropixels 高分拣结果后处理与分析的完整方案从SortingAnalyzer的创建与扩展计算、质量指标的定义与阈值筛选、波形/模板/单位位置/相关图分析到探针图、漂移图、单元摘要等可视化以及 LFP 分析、Phy/NWB/报告导出和端到端管线。读完本文你将能够独立搭建一条分拣 → 后处理 → 质量评估 → 可视化 → 导出的 Neuropixels 数据发布级分析流程并能直接使用本仓库配套的脚本与模板。从分拣结果到分析对象SortingAnalyzerSpike sorting 输出的sorting对象只包含哪个单元在哪个采样点放电这一最基本的时序信息。所有后续分析——波形、模板、噪声、PCA、相关图、质量指标——都需要把分拣结果与原始recording对齐计算这正是SortingAnalyzer的职责。它是 SpikeInterface 后处理的核心对象仓库中的 scripts/neuropixels_pipeline.py 与 assets/analysis_template.py 均围绕它组织整个后处理阶段。创建 Analyzercreate_sorting_analyzer接收分拣结果与预处理后的记录对象并以稀疏sparse方式存储只保留每个单元周围有信号的通道而不是整个探针的全部通道从而大幅压缩存储与计算量import spikeinterface.full as si # Create analyzer analyzer si.create_sorting_analyzer( sorting, recording, sparseTrue, # Use sparse representation formatbinary_folder, # Storage format folderanalyzer_output # Save location )关键参数说明sparseTrue按单元稀疏存储波形Neuropixels 1.0/2.0 拥有 384 个记录通道若全部密排存储会非常浪费format支持memory、binary_folder与zarr持久化分析建议使用binary_folder磁盘目录需要流式/大数组场景可改用 zarrfolder分析结果的落盘位置。注意run_sorter与create_sorting_analyzer使用folder参数而导出函数使用output_folder见 references/api_reference.md。计算扩展ExtensionsSortingAnalyzer的一切分析能力都通过扩展extension提供每次compute会生成一类可独立存取的分析数据# Compute all standard extensions analyzer.compute(random_spikes) # Random spike selection analyzer.compute(waveforms) # Extract waveforms analyzer.compute(templates) # Compute templates analyzer.compute(noise_levels) # Noise estimation analyzer.compute(principal_components) # PCA analyzer.compute(spike_amplitudes) # Amplitude per spike analyzer.compute(correlograms) # Auto/cross correlograms analyzer.compute(unit_locations) # Unit locations analyzer.compute(spike_locations) # Per-spike locations analyzer.compute(template_similarity) # Template similarity matrix analyzer.compute(quality_metrics) # Quality metrics # Or compute multiple at once analyzer.compute([ random_spikes, waveforms, templates, noise_levels, principal_components, spike_amplitudes, correlograms, unit_locations, quality_metrics ])扩展之间存在依赖顺序random_spikes必须先于waveforms波形从随机抽取的尖峰中截取waveforms又先于templates模板是波形的聚合统计而principal_components先于依赖 PCA 的质量指标isolation_distance、l_ratio、d_prime、nn_hit_rate。仓库脚本 scripts/neuropixels_pipeline.py 正是按此顺序分组计算的analyzer.compute(random_spikes, methoduniform, max_spikes_per_unit500) analyzer.compute(waveforms, ms_before1.5, ms_after2.0, **job_kwargs) analyzer.compute(templates, operators[average, std]) analyzer.compute(noise_levels) analyzer.compute(spike_amplitudes, **job_kwargs) analyzer.compute(correlograms, window_ms100, bin_ms1) analyzer.compute(unit_locations, methodmonopolar_triangulation) analyzer.compute(template_similarity)保存与加载# Save analyzer.save_as(folderanalyzer_saved, formatbinary_folder) # Load analyzer si.load_sorting_analyzer(analyzer_saved)已计算的扩展会被持久化加载后无需重新计算即可直接读取这是避免对几十 GB 原始数据反复重算的关键。此外analyzer.select_units(unit_ids, folder..., formatbinary_folder)可用于在筛选后生成只含合格单元的干净 analyzer对应 SKILL.md 中的analyzer_clean导出流程。质量指标量化单元纯度、完整性与稳定性质量指标是分拣结果能否进入下游分析乃至论文的通行证。仓库 references/QUALITY_METRICS.md 将指标归纳为三类问题类别回答的问题关键指标污染Type I是否混入多个神经元ISI violations、SNR完整性Type II是否漏掉了尖峰Amplitude cutoff、presence ratio稳定性单元是否随时间漂移Drift 指标、amplitude CV计算指标analyzer.compute(quality_metrics) qm analyzer.get_extension(quality_metrics).get_data() print(qm)返回的qm是 pandas DataFrame每行对应一个单元、每列对应一项指标可直接进行query筛选或.to_csv()落盘。也可以只算需要的子集analyzer.compute( quality_metrics, metric_names[snr, isi_violations_ratio, presence_ratio, firing_rate] )完整指标速查表MetricDescriptionGood ValuessnrSignal-to-noise ratio 5isi_violations_ratioISI violation ratio 0.01 (1%)isi_violations_countISI violation countLowpresence_ratioFraction of recording with spikes 0.9firing_rateSpikes per second0.1-50 Hzamplitude_cutoffEstimated missed spikes 0.1amplitude_medianMedian spike amplitude-amplitude_cvCoefficient of variation 0.5drift_ptpPeak-to-peak drift (um) 40drift_stdStandard deviation of drift 10drift_madMedian absolute deviation 10sliding_rp_violationSliding refractory period 0.05sync_spike_2Synchrony with other units 0.5isolation_distanceMahalanobis distance 20l_ratioL-ratio (isolation) 0.1d_primeDiscriminability 5nn_hit_rateNearest neighbor hit rate 0.9nn_miss_rateNearest neighbor miss rate 0.1silhouette_scoreCluster silhouette 0.5对部分指标的阈值解读可参考 references/QUALITY_METRICS.md 的分档例如isi_violations_ratio按 0.01优良单单元、0.01–0.1轻微污染、0.1–0.5疑似多单元、 0.5极可能多单元分级其生理依据是神经元约 1.5 ms 的不应期Hill et al. 2011snr按 10优秀、5–10良好、2–5可接受、 2可能是噪声分级。参数化指标计算部分指标支持自定义参数例如不应期阈值、PCA 近邻数、峰值极性等# 自定义不应期 analyzer.compute(quality_metrics, metric_names[isi_violations_ratio], isi_threshold_ms1.5, min_isi_ms0.0) # PCA 空间隔离指标 analyzer.compute(quality_metrics, metric_names[isolation_distance], n_neighbors4) # 幅度截断的峰值极性 analyzer.compute(quality_metrics, metric_names[amplitude_cutoff], peak_signneg) # neg, pos, or both # 存在率的时间窗 analyzer.compute(quality_metrics, metric_names[presence_ratio], bin_duration_s60) # 1-minute bins # 漂移指标 analyzer.compute(quality_metrics, metric_names[drift_ptp, drift_std, drift_mad])依赖 PCA 的隔离类指标isolation_distance、l_ratio、d_prime、nn_hit_rate、silhouette_score需先计算principal_components默认n_components5。一次性计算全部指标时可按 references/QUALITY_METRICS.md 中的all_metric_names列表组织放电属性firing_rate、presence_ratio、波形snr、amplitude_cutoff、amplitude_cv_median、amplitude_cv_range、ISIisi_violations_ratio、isi_violations_count、漂移drift_ptp、drift_std、drift_mad、隔离PCA 类、同步sync_spike_2/4/8。自定义质量阈值筛选拿到指标表后即可定义自己的质量标准并筛选合格单元qm analyzer.get_extension(quality_metrics).get_data() # Define quality criteria quality_criteria { snr: (, 5), isi_violations_ratio: (, 0.01), presence_ratio: (, 0.9), firing_rate: (, 0.1), amplitude_cutoff: (, 0.1), } # Filter good units good_units qm.query( (snr 5) (isi_violations_ratio 0.01) (presence_ratio 0.9) ).index.tolist() print(fGood units: {len(good_units)}/{len(qm)})业界常用三套标准详见 references/QUALITY_METRICS.mdAllen Institute宽松presence_ratio 0.95 and isi_violations_ratio 0.5 and amplitude_cutoff 0.1IBL严格presence_ratio 0.9 and isi_violations_ratio 0.1 and amplitude_cutoff 0.1 and firing_rate 0.1Strict 单单元用于精确时序分析snr 5 and presence_ratio 0.99 and isi_violations_ratio 0.01 and amplitude_cutoff 0.01 and isolation_distance 20 and drift_ptp 40。仓库 scripts/compute_metrics.py 将这三套标准固化为CURATION_CRITERIA预设注意该脚本中snr依次为 3.0 / 4.0 / 5.0isi_violations_ratio依次为 0.5 / 0.1 / 0.01并对每个单元输出good/mua/noise标签tests/neuropixels-analysis/test_scripts.py 中的测试用例验证了能通过所有预设的单元必须是一个真正优秀单元EXEMPLARY以及strict 至少与 ibl 一样严格、ibl 至少与 allen 一样严格这一单调性约束防止阈值漂移造成静默的科学错误。波形与模板分析波形是单元信号的基本特征模板则是波形的统计聚合用于比较、分类与可视化。提取波形analyzer.compute(waveforms, ms_before1.5, ms_after2.5, max_spikes_per_unit500) # Get waveforms for a unit waveforms analyzer.get_extension(waveforms).get_waveforms(unit_id0) print(fShape: {waveforms.shape}) # (n_spikes, n_samples, n_channels)ms_before/ms_after决定截取窗口max_spikes_per_unit限制每个单元抽取的尖峰数默认约 500 足以稳定估计波形同时控制计算量。返回数组形状为(n_spikes, n_samples, n_channels)。计算模板analyzer.compute(templates, operators[average, std, median]) # Get template templates_ext analyzer.get_extension(templates) template templates_ext.get_unit_template(unit_id0, operatoraverage)operators可同时计算average、std、median等多种聚合std模板可直接用于绘制均值 ± 标准差的误差带references/plotting_guide.md 中的多面板单元摘要即用get_unit_template(unit_id, operatorstd)填充置信区间。模板相似度analyzer.compute(template_similarity) sim analyzer.get_extension(template_similarity).get_data() # Matrix of cosine similarities between templates该扩展输出单元模板两两之间的余弦相似度矩阵用于识别可能重复分拣的单元对或高度相似的模板。单元位置分析计算位置analyzer.compute(unit_locations, methodmonopolar_triangulation) locations analyzer.get_extension(unit_locations).get_data() print(locations) # x, y coordinates per unit尖峰位置analyzer.compute(spike_locations, methodcenter_of_mass) spike_locs analyzer.get_extension(spike_locations).get_data()位置方法选择center_of_mass快速、精度较低monopolar_triangulation更精确、较慢grid_convolution精度与速度的平衡选择。unit_locations给出每个单元在探针平面上的 x、y 坐标可叠加在探针图上查看单元的空间分布spike_locations则给出单个尖峰的位置可用于绘制尖峰位置散布图见 references/plotting_guide.md 的plot_spike_locations用法。相关图分析analyzer.compute(correlograms, window_ms50, bin_ms1) correlograms, bins analyzer.get_extension(correlograms).get_data() # correlograms shape: (n_units, n_units, n_bins) # Auto-correlogram for unit i: correlograms[i, i, :] # Cross-correlogram units i,j: correlograms[i, j, :]相关图是三维张量(n_units, n_units, n_bins)。对角切片[i, i, :]是单元 i 的自相关图autocorrelogram用于检查不应期凹槽refractory dip判断单元是否干净非对角切片[i, j, :]是单元对互相关图cross-correlogram可揭示单元间的功能连接或共享噪声源。window_ms为分析窗口宽度bin_ms为直方图分辨率。可视化从探针图到单元摘要SpikeInterface 的绘图 API 统一挂载在si命名空间下底层为spikeinterface.widgets as sw见 references/api_reference.md一行调用即可产出出版级图表。仓库 references/plotting_guide.md 还提供了完整的 Matplotlib 出版配置dpi、字体、单栏/双栏尺寸、色盲友好配色。探针与单元模板si.plot_probe_map(recording, with_channel_idsTrue) # All units si.plot_unit_templates(analyzer) # Specific units si.plot_unit_templates(analyzer, unit_ids[0, 1, 2])波形# Plot waveforms with template si.plot_unit_waveforms(analyzer, unit_ids[0]) # Waveform density si.plot_unit_waveforms_density_map(analyzer, unit_id0)Raster 图与幅度si.plot_rasters(sorting, time_range(0, 10)) # First 10 seconds analyzer.compute(spike_amplitudes) si.plot_amplitudes(analyzer) # Distribution si.plot_all_amplitudes_distributions(analyzer)相关图# Auto-correlograms si.plot_autocorrelograms(analyzer, unit_ids[0, 1, 2]) # Cross-correlograms si.plot_crosscorrelograms(analyzer, unit_ids[0, 1])质量指标可视化# Summary plot si.plot_quality_metrics(analyzer) # Specific metric distribution import matplotlib.pyplot as plt qm analyzer.get_extension(quality_metrics).get_data() plt.hist(qm[snr], bins50) plt.xlabel(SNR) plt.ylabel(Count)references/QUALITY_METRICS.md 还给出了多指标直方图矩阵2×3 子图、叠加阈值虚线以及发放率 vs SNR散点以 ISI 违例率为颜色映射等进阶方案用于批量审视所有单元的质量分布。位置、漂移与单元摘要si.plot_unit_locations(analyzer) si.plot_drift_raster(sorting, recording) # Comprehensive unit summary si.plot_unit_summary(analyzer, unit_id0)plot_unit_summary一张图集成波形、模板、自相关图、幅度、位置等关键信息是 AI 辅助审核与人工复核单元的最快入口仓库 references/AI_CURATION.md 描述的渲染单元摘要图 → 交由 Agent 评估隔离质量模式正依赖它。LFP 分析除尖峰外Neuropixels 的.lf.bin流低频带还提供局部场电位LFP可用于振荡、频谱与状态分析。加载 LFPlfp si.read_spikeglx(/path/to/data, stream_nameimec0.lf) print(fLFP: {lfp.get_sampling_frequency()} Hz)Neuropixels 1.0 的 AP 流采样率为 30 kHzLF 流为 2.5 kHz二者由同一.meta文件描述通过stream_name选择si.get_neo_streams(spikeglx, path)可先列出imec0.ap/imec0.lf/nidq等流见 references/api_reference.md。基础处理# Downsample if needed lfp_ds si.resample(lfp, resample_rate1000) # Common average reference lfp_car si.common_reference(lfp_ds, referenceglobal, operatormedian)降采样降低后续频谱计算成本全局中位参考可去除共同噪声源。提取波形与频谱分析import numpy as np # Get traces (channels x samples) traces lfp.get_traces(start_frame0, end_frame30000) # Specific channels traces lfp.get_traces(channel_ids[0, 1, 2])from scipy import signal import matplotlib.pyplot as plt # Get single channel trace lfp.get_traces(channel_ids[0]).flatten() fs lfp.get_sampling_frequency() # Power spectrum freqs, psd signal.welch(trace, fs, nperseg4096) plt.semilogy(freqs, psd) plt.xlabel(Frequency (Hz)) plt.ylabel(Power) plt.xlim(0, 100)频谱图f, t, Sxx signal.spectrogram(trace, fs, nperseg2048, noverlap1024) plt.pcolormesh(t, f, 10*np.log10(Sxx), shadinggouraud) plt.ylabel(Frequency (Hz)) plt.xlabel(Time (s)) plt.ylim(0, 100) plt.colorbar(labelPower (dB))结果导出Phy、NWB 与报告导出到 Phy人工审核si.export_to_phy( analyzer, output_folderphy_export, compute_pc_featuresTrue, compute_amplitudesTrue, copy_binaryTrue ) # Then: phy template-gui phy_export/params.pyPhy 是电生理社区标准的人工审核工具compute_pc_featuresTrue与compute_amplitudesTrue提供特征空间供人工聚类修正。注意export_to_phy使用output_folder参数与run_sorter的folder不同参见 references/api_reference.md 的特别说明。导出到 NWB数据共享from spikeinterface.exporters import export_to_nwb export_to_nwb( recording, sorting, output.nwb, metadatadict( session_descriptionNeuropixels recording, experimenterName, labLab name, institutionInstitution ) )NWB 是神经数据标准格式附带的 metadata 会写入文件头部保证实验溯源信息完整。导出报告si.export_report( analyzer, output_folderreport, remove_if_existsTrue, formathtml )format支持html与png仓库 scripts/neuropixels_pipeline.py 使用formatpng生成可嵌入论文的图集。完整分析管线一个可复用的函数将以上所有环节串起来即构成一条发布级管线。references/ANALYSIS.md 提供了完整实现import spikeinterface.full as si def analyze_sorting(recording, sorting, output_dir): Complete post-processing pipeline. # Create analyzer analyzer si.create_sorting_analyzer( sorting, recording, sparseTrue, folderf{output_dir}/analyzer ) # Compute all extensions print(Computing extensions...) analyzer.compute([random_spikes, waveforms, templates, noise_levels]) analyzer.compute([principal_components, spike_amplitudes]) analyzer.compute([correlograms, unit_locations, template_similarity]) analyzer.compute(quality_metrics) # Get quality metrics qm analyzer.get_extension(quality_metrics).get_data() # Filter good units good_units qm.query( (snr 5) (isi_violations_ratio 0.01) (presence_ratio 0.9) ).index.tolist() print(fQuality filtering: {len(good_units)}/{len(qm)} units passed) # Export si.export_to_phy(analyzer, f{output_dir}/phy) si.export_report(analyzer, f{output_dir}/report) # Save metrics qm.to_csv(f{output_dir}/quality_metrics.csv) return analyzer, qm, good_units # Usage analyzer, qm, good_units analyze_sorting(recording, sorting, output/)仓库现成脚本无需从零搭建本仓库的 scripts 目录提供了完整工具链explore_recording.py快速检查记录流、通道、时长、坏道preprocess_recording.py自动化预处理run_sorting.py运行分拣器compute_metrics.py计算质量指标并按allen/ibl/strict预设筛选对应 SKILL.md 的 curation 步骤export_to_phy.py导出 Phy 供人工审核neuropixels_pipeline.py端到端一键管线python scripts/neuropixels_pipeline.py /path/to/spikeglx/data output/ --sorter kilosort4 --curation allenscripts/neuropixels_pipeline.py 的run_pipeline会依次执行数据加载自动识别 SpikeGLX/Open Ephys→ 预处理高通滤波、坏道去除、相位校正、公共中位参考→ 漂移检测峰检测 质心定位drift_estimate超过 20 μm 才触发运动校正→ 分拣 → 后处理 → 质量指标 → 阈值 curation → 导出Phy、PNG 报告、quality_metrics.csv、unit_labels.json、summary.json。其命令行支持--sorterkilosort4/kilosort3/spykingcircus2/mountainsort5、--stream、--no-motion-correction、--curation等选项。若需要完全可编辑的脚本assets/analysis_template.py 将全部参数集中在文件顶部的PARAMETERS区DATA_PATH、DATA_FORMAT、FREQ_MIN/FREQ_MAX、APPLY_PHASE_SHIFT、MOTION_PRESET、SORTER_PARAMS、CURATION_METHOD、N_JOBS等复制后改参数即可运行cp assets/analysis_template.py my_analysis.py # Edit the PARAMETERS section, then run python my_analysis.py最佳实践要点结合 SKILL.md 的常见陷阱清单实战中应注意分拣前务必检查漂移——漂移超过约 10 μm 会显著降低质量先detect_peakslocalize_peaks绘制漂移光栅图需要时再用correct_motion(rec, presetnonrigid_fast_and_accurate, foldermotion/)校正Neuropixels 1.0 必须做phase_shift修正 ADC 采样偏移保存预处理后的记录rec.save(folder..., formatbinary)避免重复计算Kilosort 也需要二进制文件Kilosort4 建议使用 GPUCPU 场景可选 SpykingCircus2、Mountainsort5、Tridesclous2自动化/模型化筛选只是起点边界单元应结合人工或 AI 复核仓库提供 references/AUTOMATED_CURATION.md 与 references/AI_CURATION.md 两套进阶方案记录阈值与模型 repo ID保证结果可复现关键实验导出到 Phy 保留人工监督。依赖安装Python ≥ 3.10推荐 uv参考 SKILL.md 的 Installation 章节核心为uv pip install spikeinterface[full] probeinterface neo分拣器按需追加kilosort、spykingcircus、mountainsort5模型化筛选需huggingface_hub与skopsAI 审核可选anthropic。正式管线建议锁定版本文档示例spikeinterface0.104.3、kilosort4.1.7、probeinterface0.3.2、neo0.14.4。深入阅读完整工作流references/standard_workflow.mdAPI 速查references/api_reference.md出版级绘图配方references/plotting_guide.md预处理细节references/PREPROCESSING.md分拣细节references/SPIKE_SORTING.md运动校正references/MOTION_CORRECTION.md质量指标深入references/QUALITY_METRICS.md自动化与模型化筛选references/AUTOMATED_CURATION.mdAI 辅助审核references/AI_CURATION.md【免费下载链接】scientific-agent-skillsTurn any AI agent into an AI Scientist. The #1 Agent Skills library for science, used by 190,000 scientists worldwide. 165 ready-to-use validated skills plus 100 scientific databases covering biology, chemistry, medicine, and drug discovery. Compatible with Cursor, Claude Code, Codex, Pi, Antigravity, and the open Agent Skills standard.项目地址: https://gitcode.com/GitHub_Trending/cl/scientific-agent-skills创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考
