Flowmeris
Flow cytometry analysis that runs entirely in the browser. You load a folder of FCS files as a group,
gate them with one shared hierarchy (with per-sample adjustments), plot them, and export statistics,
plots and Gating-ML. Every method is documented, and every result is checked against reference
implementations.
- Input: FCS 2.0 / 3.0 / 3.1 / 3.2, including multi-dataset files. Compensation comes from each
sample's
$SPILLOVER keyword, an edited matrix, or an imported CSV.
- Scales: linear, log10, logicle, arcsinh and hyperlog, using the Gating-ML 2.0 definitions and
editable parameters.
- Plots: dot, pseudocolor, density, contour (equal-probability or logarithmic levels, with
outliers) and histogram; tiled per-sample panels and ridge plots for groups.
- Gates: rectangle, ellipse, polygon, quadrant (with offset legs), curly quadrant, spider, histogram
range and bisector. Gates are hierarchical, you can drill into any population, and edits apply to the
whole group or to one sample only (overrides are flagged everywhere).
- Output: tidy and wide statistics CSV with provenance, Gating-ML 2.0, gated events as FCS 3.1 or
CSV, SVG/PNG plots at 300 dpi, and the workspace as JSON.
- Privacy: no server. Data stay in the browser's private storage. A Content-Security-Policy forbids
network requests to anywhere but the app's own origin.
Validation
| Check | Result |
|---|
| ISAC Gating-ML 2.0 compliance suite (51 reference gates) | exact event-level agreement |
| FCS parsing, compensation and transforms vs FlowKit 1.2.3 (pinned) | ≤ 10⁻¹⁴, 10⁻⁹ and 10⁻¹⁰ relative |
| Statistics vs NumPy, on gated populations in linear and transformed units | ≤ 10⁻¹² relative (transformed: 10⁻¹⁰); percentiles exact on linear channels |
| Replicate summaries and t-based 95% CI vs SciPy/pandas | ≤ 10⁻¹² (t distribution 10⁻¹⁰) |
| Gating-ML exported by the app, evaluated by FlowKit (every gate kind) | identical membership, event by event |
| Gates drawn in the app, exported as Gating-ML and evaluated by FlowKit | identical population counts |
Details: docs/validation. Methods: docs/methods.
Development
Requires Node ≥ 20 (pnpm via corepack). To regenerate the golden fixtures you also need
uv.
corepack pnpm install
corepack pnpm fixtures:fetch # large test files, SHA-256 verified
corepack pnpm dev # docs at http://localhost:5173, app at /app/
corepack pnpm check # lint + typecheck + unit/golden tests
corepack pnpm build && corepack pnpm e2e # Playwright (Chromium, Firefox, WebKit)
corepack pnpm golden # regenerate fixtures/golden from FlowKit (pinned in uv.lock)
corepack pnpm build:site && corepack pnpm preview # docs + app (at /app/), as deployed
Layout
Notes for coding agents (layering, tests, conventions) are in CLAUDE.md.
| Path | Contents |
|---|
packages/model | workspace schema (zod), canonical JSON, fingerprints |
packages/fcs | FCS parser and writer |
packages/transforms, compensation, gating, stats, density, render | numerical kernels |
packages/gatingml, export | Gating-ML reader/evaluator; CSV and Gating-ML export |
packages/table | statistics table: derived columns, formulas, replicate summaries, table import |
packages/testkit | test-only helpers: fixture paths, tolerances, synthetic data |
packages/engine, storage | cached analysis pipeline; OPFS persistence |
apps/web | React app and compute workers |
docs | VitePress site: guide, methods, validation, decision records |
fixtures, tools/golden | reference data and the golden-value generator (fixtures/golden/inputs: files the app writes, evaluated by FlowKit) |
Status
Milestones M0–M3 of the implementation plan are complete, along with parts of M4–M5 (tiles, ridge
plots, spider gates, Gating-ML/FCS/SVG/PNG export). Not yet implemented:
- XLSX export;
- PDF plot export;
- N×N plot matrix;
- Time-channel QC view;
- Boolean gates in the UI;
- compensation from single-stain controls;
- Gating-ML import in the UI;
- WebAssembly kernels.
License
Not yet chosen. Test fixtures keep their upstream licenses (see fixtures/PROVENANCE.md).
