Foundation models,for neuroscience data.
Neuron Lab pairs classical, reproducible pipelines with Neuron-FM — cell, signal and connectome foundation models — running on your data with real math on the server.
168 pipelines, real math, tested end-to-end.
Every module runs a genuine algorithm on the server — Welch ANOVA, Games-Howell post-hoc, Hedges' g, empirical-Bayes moderation, Huber regression, permutation tests and bootstrap CIs — backed by unit tests, not placeholder charts.
Differential expression
Welch's t, limma-style empirical-Bayes moderation, BH-FDR — volcano + top-genes table.
PCA & clustering
Covariance-eigendecomposition PCA with scree, projections and explained variance.
Spike-train stats
Firing rate, ISI, CV, jittered cross-correlograms and population dynamics.
LFP / EEG spectra
Welch and multitaper PSD, band power, spectral entropy, Hjorth parameters.
HH / LIF neurons
RK4 integration of Hodgkin–Huxley, LIF, AdEx, Izhikevich with spike detection.
GWAS association
Case/control scans with genomic control λ, BH q-values and Manhattan plots.
Group comparisons
Welch ANOVA, Kruskal–Wallis, Games-Howell post-hoc, Hedges' g with 95% CI.
Robust & resampling
Huber M-estimator regression, permutation tests and percentile bootstrap CIs.
168 modules
One launcher, one result viewer — from survival and connectome metrics to eQTL scans.
Foundation models, integrated end-to-end.
Cell, signal and connectome FMs share the same launcher, storage, RLS and result viewer as every classical analysis — swap the backbone, keep the pipeline.
Cell Atlas FM
PCA embedding · scGPT / Geneformer-class in production
Embed cells, then annotate cell types zero-shot or cluster them — accuracy on held-out cells.
Neural Signal FM
spectral + PCA · LaBraM / CBraMod-class in production
Embed neural-signal windows and classify dynamical state — regular / bursting / irregular.
Connectome FM
spectral node embedding · graph pretraining in production
Embed connectome nodes and detect communities; ARI vs planted modules.
Built for research that has to be reproducible.
Every dataset and analysis is owned by a user, protected by row-level security, and versioned by parameters — so a result from today can be replayed tomorrow.
Per-user isolation
Row-level security scopes every dataset, analysis, and result to its owner.
Dataset storage
Signed uploads, private storage, grouped by project.
Rich viewers
Metrics grids, sortable tables, volcano/scatter/line charts, downloads.
Reproducible
Typed parameters, deterministic runs, versioned modules.
Ready to run your first analysis?
Create an account, upload a dataset, and launch a real pipeline in minutes.