Biosensor

A lightweight Python package with an attached local viewer that converts raw electrochemical instrument exports into one tidy dataframe schema, and lets you visually verify the parse before trusting the output.

Targets solo researchers and small labs working on immunosensor and biosensor cyclic voltammetry (CV) / DPV / SWV data who need a fast, local path from raw instrument export to usable data, without adopting an institutional ELN.

Supported formats (v1)

• CH Instruments text export
• Metrohm Nova (Autolab) text/CSV export
• PalmSens .pssession (best-effort; see src/biosensor/readers/palmsens.py)
• Generic delimited CSV: a named header, metadata lines before the header, or headerless two-column numeric (potential, current)

Format detection is content-based (file signature and header content), not filename-extension based.

Install

The library, from PyPI:

pip install biosensor

The viewer, launchers, and sample_data/ ship in the repository, not the wheel. For those and the test suite, clone and install in place:

python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

Library usage

from biosensor import load, batch_load, to_dataframe

result = load("cv01.txt")
df = to_dataframe(result.measurement)
print(result.qc.sanity_status)  # "ok" | "flagged" | "failed"

batch = batch_load("data/2026-08-05-run/")
df = batch.to_dataframe()          # all files, one tidy frame
qc_df = batch.qc_dataframe()       # per-file sanity status

Every reader converts its instrument-specific format into a Measurement (potential, current, scan rate, cycle number, technique, sample ID, analyte concentration, and a flexible technique_params dict for things like SWV frequency or DPV pulse width). QC state (sanity_status, review notes) lives in a separate QCRecord so the sanity heuristic can evolve without touching the measurement schema.

Viewer

Double-click the launcher for your platform (launcher.command on macOS, launcher.sh on Linux, launcher.bat on Windows). Each sets up the venv on first run, then starts the server and opens the browser:

./launcher.command      # macOS
./launcher.sh           # Linux
launcher.bat            # Windows
# or manually: source .venv/bin/activate && python viewer/app.py

Opens at http://127.0.0.1:5050. Three panes: file list on the left (live filter, ok/flagged/failed tabs), the CV curve (Plotly, zoom/pan) with a Dataframe tab in the center, and a parse record on the right (quality check, instrument metadata, sample/concentration mapping). Load a single file or an entire folder, correct a wrong column mapping in-place (see a live preview before applying), override the quality flag manually, and export any file or the whole batch as CSV. Light/dark theme toggle in Settings. htmx, Plotly, and the IBM Plex fonts are vendored under viewer/static/vendor/, so the viewer runs fully offline with no CDN dependency. (Fonts are IBM Plex, under the SIL Open Font License 1.1; see viewer/static/vendor/fonts/OFL.txt.)

Visual design follows the Tree Design System, with design tokens under viewer/static/tokens/.

The viewer is local, single-user, in-memory only; state resets on restart. This is intentional (see Non-goals below).

Tracing

Action-level tracing via traceact (file parse, batch upload, mapping correction, CSV export) writes to data/traces/traces.jsonl for local debugging. It isn't required to run the app, and it's gitignored.

Sanity-check heuristic (v1)

A simple curve-shape check, not anything trained: flags constant/flat current or potential (near-certain wrong column mapping), non-finite values, too few points, a single-direction sweep with no return cycle, and the absence of any peak/inflection in the current trace. Manual override is always available in the viewer.

Security

The viewer accepts arbitrary uploaded files and treats them as untrusted input:

• Format detection inspects file content, never trusts the extension alone
• Per-file byte and data-row limits bound parsing cost (see readers/base.py)
• Any parse failure, including malformed or adversarial files, degrades to a per-file error, never a server crash
• Filenames, sample IDs, and technique strings derived from file content are sanitized before display and CSV export (including a CSV-formula injection guard)
• No macro or script execution in any format reader

Non-goals (v1)

Peak fitting, baseline correction, calibration curves, ML modeling, multi-user access, authentication, or cloud deployment. Four format readers is the v1 target, not exhaustive vendor coverage.

Sample data

sample_data/ has a synthetic ~27-file batch (all four formats) for exercising the viewer by hand: an IL6 immunosensor concentration series (CH Instruments, peak height scales with concentration), ferricyanide and blank references, Metrohm Nova DPV/CV, PalmSens sessions, a well-formed generic CSV, and two files that deliberately fail to parse (to exercise the batch error path). Point "Load a folder" at it. Regenerate with:

source .venv/bin/activate
python scripts/generate_sample_data.py

Tests

source .venv/bin/activate
pytest

Fixtures in tests/fixtures/ are synthetic (generated, not instrument exports) but shaped like each format's structure, including one unrecognizable file to exercise the error path.

Documentation

• USAGE.md: the full manual (API, schema, formats, errors)
• ARCHITECTURE.md: how the library and viewer fit together
• CHANGELOG.md: dated changes per version

Project layout

src/biosensor/       # the library: schema, readers, core API, QC heuristic
viewer/                 # Flask + HTMX + Plotly local viewer
tests/                  # pytest suite + synthetic fixtures

Author

Built by Mo Shehu, mohammedshehu.com