Neural Population Analysis Guide (Claude Skill)

Guides Claude through latent-variable analysis of simultaneously recorded neural populations — picking between PCA, GPFA, dPCA and jPCA, normalizing spike rates correctly, and choosing the number of dimensions by cross-validation rather than a variance rule of thumb.

   
Type Claude Skill
Supplier Awesome Cognitive and Neuroscience Skills (community OSS, MIT)
Availability GA — one of ~40 research skills in the collection
Pricing Free / OSS (MIT)
Capabilities Read/Write — methodology guidance; Claude writes and runs the analysis code locally
Verified works · 2026-08-13
Security cleared · 2026-08-13 — genuine org transfer to NeuroAIHub confirmed, MIT, no external credentials

How to install

  • Claude Code — plugin marketplace (installs all skills in the collection): 
    /plugin marketplace add HaoxuanLiTHUAI/awesome_cognitive_and_neuroscience_skills
/plugin install awesome-cognitive-and-neuroscience-skills@awesome-cognitive-and-neuroscience-skills

    Restart Claude Code afterwards. The skills are description-activated — there is no slash command; ask a population-analysis question and Claude loads the skill.

  • Claude Code — single-skill alternative. This skill declares a required dependency on the collection’s research-literacy skill, so copy both: 
    git clone https://github.com/HaoxuanLiTHUAI/awesome_cognitive_and_neuroscience_skills
cp -r awesome_cognitive_and_neuroscience_skills/skills/neural-population-analysis-guide ~/.claude/skills/
cp -r awesome_cognitive_and_neuroscience_skills/skills/research-literacy ~/.claude/skills/

    (Project-scoped alternative: copy into .claude/skills/ instead. The repo’s default branch is master, not main.)

  • Underlying software — the skill is method guidance, not an analysis engine. It names no pinned packages; install whichever implementation you select (for example pip install elephant for GPFA, the Kobak dPCA reference implementation, or scikit-learn for PCA).

What it does

Runs a four-stage protocol — research planning, method selection, preprocessing and analysis, then validation and reporting.

Method selection decision tree:

  • Dominant co-variation patterns across neurons → PCA (linear, static)
  • Smooth single-trial latent trajectories → GPFA or factor analysis
  • Variance attributable to specific task parameters → dPCA (demixes stimulus, decision and time)
  • Rotational dynamics in motor-cortex-style data → jPCA
  • Visualization only → t-SNE/UMAP, explicitly not for quantitative claims

Recording requirements the skill states per method: PCA needs 30+ neurons; GPFA needs 50–100 neurons, 20–50 ms bins and 50+ trials per condition; dPCA needs 20–30+ neurons and 20+ trials per condition.

Dimensionality selection: parallel analysis (compare eigenvalues against shuffled data) or cross-validated reconstruction error. GPFA latent count is chosen by leave-one-neuron-out cross-validated log-likelihood over 2–15 dimensions; dPCA regularization lambda is chosen on held-out trials.

Primary use cases: Neuropixels and multi-electrode array population dynamics, motor and decision-making trajectory analysis, task-variable demixing.

Notes

AI-generated content — verify before use. All skills in this collection carry review_status: ai-generated, and the README states the content “has not been individually verified by human domain experts.” This skill cites Cunningham & Yu 2014, Yu et al. 2009, Kobak et al. 2016, Churchland et al. 2012, King & Dehaene 2014 and Humphries 2021 — check the neuron/trial minima and normalization constants against those sources.

Two rules the skill treats as non-negotiable, both worth knowing before you install it:

  • Do not z-score firing rates. Standard z-scoring inflates low-firing neurons into apparent structure. Use soft normalization — (rate − mean) / (range + constant) with a constant of 5 spikes/s — or a square-root transform.
  • There is no 90% variance rule. The skill explicitly rejects the standard data-science “keep enough PCs for 90% of variance” heuristic as meaningless for neural data; use parallel analysis or cross-validation instead.

Related catalogued skills in the same collection: Parameter Recovery Checker for validating that a fitted latent model is identifiable, Calcium Imaging Analysis Guide for the upstream extraction step when the population comes from imaging rather than electrophysiology, and Drift-Diffusion Model for the behavioural counterpart. For the recording-side pipeline see SpikeInterface.

Sources

Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.