Convert raw analytical instrument data to Allotrope ASM JSON

Hand Claude Code a vendor-format file from a cell counter, plate reader, HPLC, mass spectrometer, or qPCR instrument; get back a strict-validated Allotrope Simple Model (ASM) JSON, a flattened 2D CSV ready for LIMS ingestion, and an exportable Python parser that a data engineer can put behind a pipeline.

   
Problem class Workflow automation
Subject areas Chemistry, Drug Repurposing and Discovery, Molecular and Cellular Biology, Translational Medicine
Evidence level Reported
Complexity One skill or MCP
Availability Fully open
Compute Laptop

Problem

Every analytical instrument writes its own file format — Beckman Vi-CELL counters emit one CSV layout; Agilent HPLCs emit another; Tecan and Molecular Devices plate readers each emit their own XML / Excel hybrids; NanoDrop spreadsheets, BMG OMEGA exports, Roche LightCycler .ixo packages, Bio-Rad CFX .pcrd. A pharma or biotech lab pipes those into LIMS, an ELN, or a data lake — and the glue is a long tail of one-off parsers that drift as vendors push firmware updates. The Allotrope Foundation standardised this with the Allotrope Simple Model (ASM): a JSON-LD schema per technique (flow cytometry, plate-reader, HPLC-DAD, qPCR, …) with ontology-anchored field names and explicit raw-vs-derived data separation.

The work that has to happen on every new file: identify the instrument and vendor, pick the right parser, map fields into the right ASM section (raw measurements into measurement-document; derived values into calculated-data-aggregate-document with data-source-aggregate-document traceability), validate against the ASM schema, and emit either ASM JSON (for an ASM-aware sink) or a flattened CSV (for everything else). Solved looks like: drop the file, get JSON + CSV + a re-usable parser script, no hand editing.

  1. Install the instrument-data-to-allotrope skill — bundled in the Anthropic life-sciences marketplace and also inside the bio-research plugin:

    /plugin marketplace add anthropics/life-sciences
/plugin install instrument-data-to-allotrope@life-sciences
pip install allotropy

    Confirm with /plugin list. The skill bundles four scripts — convert_to_asm.py, flatten_asm.py, export_parser.py, validate_asm.py — plus references for the 40+ supported instruments.

  2. Drop the raw file and ask for auto-detection. A minimal prompt:

    Convert raw/viCell_Results.csv to Allotrope ASM JSON. Auto-detect
the instrument, report your detection confidence, run the native
`allotropy` parser, and write:
  out/viCell_Results.asm.json   (full ASM JSON-LD)
  out/viCell_Results.flat.csv   (flattened 2D form for LIMS)
Then run strict ASM schema validation on the JSON and print the
validation result (pass / list of failed fields).

    The skill detects the vendor signature from the file header and reports a confidence score (the published Vi-CELL example reaches 95%). If confidence is low or the instrument is unsupported, the skill falls back to the flexible / PDF-table parser — flag this in your downstream metadata because field completeness drops.

  3. Pin the vendor explicitly for batch jobs. Auto-detection is convenient for one-off files; for a directory of 50 plate-reader exports, pass the vendor enum directly so you do not pay the detection cost 50 times:

    For every *.xlsx under raw/plate_runs/2026-Q2/, call
allotrope_from_file(path, Vendor.MOLDEV_SOFTMAX_PRO) and
write the ASM JSON next to the source file (.asm.json suffix).
On any parse failure, append a row to logs/parse_failures.csv with
path, vendor, error class, error message. Print a final summary:
N parsed / N failed.

    Vendor enums match the names in the allotropy.parser_factory module; see supported_instruments.md in the skill bundle for the canonical list.

  4. Generate a stand-alone parser for production. When a workflow stabilises, ask the skill to emit a self-contained script your data engineer can deploy without the skill in the loop:

    Generate a stand-alone Python parser for Beckman Vi-CELL BLU files.
- Input: a Vi-CELL CSV path
- Output: ASM JSON + flattened CSV, written next to the input
- Include sample input and expected output as docstrings
- Document all assumptions and the allotropy version used
- Save to pipelines/parsers/vicell_blu_parser.py
Also write a parallel notebook version at
notebooks/vicell_blu_parser.ipynb that loads a sample file and
shows the ASM structure.

  5. Validate the raw-vs-derived split before shipping. The ASM model’s single most-violated rule is mixing raw measurements with calculated values in the same document. Confirm before downstream ingestion:

    Run validate_asm on out/viCell_Results.asm.json with strict=True.
Print any field that is in `calculated-data-aggregate-document`
but lacks a `data-source-aggregate-document` traceability link,
and any field in `measurement-document` that looks like a derived
value (units of %, ratios, normalised values).

Why this assembly

Rung 2 of the simplicity ladder. The work is mechanical — auto-detect vendor, dispatch to the right allotropy parser, emit JSON and CSV, validate — but every step needs vendor-specific knowledge of where fields live and the ASM-vs-vendor field-name map. The Anthropic-shipped skill bundles 40+ instrument mappings, the strict-validator, and the raw/derived field-classification reference. Plain Claude Code can write a parser from allotropy source, but it re-derives the field mapping every session and skips the ASM-validation step by default. A multi-tool harness or autonomous system adds nothing — there is no scientific decision to make, only a format transform.

Availability

Fully open. The skill and its supporting allotropy Python library are MIT-licensed. The Allotrope Simple Model schemas themselves are open, ontology-anchored (Allotrope Foundation Ontologies, BAO, UO, IAO). No subscription, no cloud account; everything runs locally against your raw files.

Compute requirements

Laptop. Per-file parse times are sub-second for typical instrument outputs (≤10 MB CSV / Excel). The strict-validator and the JSON-Schema check are similarly fast. RAM is the file size plus a few megabytes for the ASM in-memory representation. The only step that benefits from parallelism is batch conversion of large directories — use n_jobs=-1 in the prompt.

Evidence

Reported. The skill is shipped and documented by Anthropic as part of the Claude for Life Sciences launch (October 2025) and is bundled inside the bio-research plugin. The Anthropic Getting Started with Claude for Life Sciences tutorial walks through a worked Vi-CELL example end-to-end (detect → parse → flatten → validate → emit parser script), and the GUVI step-by-step guide reproduces the workflow on a public plate-reader CSV.

Underneath the skill, the allotropy library that does the parsing is the Benchling Open Source reference implementation that the Allotrope Foundation cites in its ASM documentation; it is in production use across pharma data engineering, with 40+ vendor parsers maintained as of 2026. No peer-reviewed head-to-head benchmark of “Claude + this skill” against a hand-written parser is known — the agent loop adds discoverability, validation, and the emitted-parser handoff, not new parsing capability.

Alternatives considered

  • Plain Claude Code + allotropy library, no skill. Works for one-off files where you already know the vendor. Reach for it when a single CSV needs converting and you can name the Vendor enum yourself. The skill’s value is the auto-detect, the strict-validation, and the parser-export step.
  • Vendor LIMS connectors (Benchling, LabVantage, Sapio). Closed-source and tied to a LIMS contract. Use them when your sink is the same vendor’s LIMS and you have the licence; the skill’s value is the agent-driven ad-hoc workflow when no vendor connector exists or when you need to migrate between LIMS systems.
  • A hand-written Python parser per instrument. Maintainable but a permanent maintenance debt as vendors push firmware updates. Pin to the skill’s export_parser.py output and version-control the generated script — re-generate it when the skill bumps allotropy versions.

See also

Sources

Tried this recipe?

Share feedback — what worked, what didn’t, what you’d change. The form opens with this recipe pre-selected and a link back to this page.