Typed Parquet exports

Carry native clonotype and rearrangement evidence into columnar analysis tools.

ClonoLoom 0.9.0. Use a matching host and clients (API 2, catalog 4, policy 1).

Choose a native table

In Reports → Typed Parquet exports, choose clonotypes or rearrangements and select Export Parquet ZIP. The Clonotypes export menu offers the same choices. The active definition, QC and invariant-species setting apply. The ZIP contains the table, a schema, provider notice and hashed inventory. Original identifiers and source provenance are retained.

Native Parquet export receipt for 46,447 rearrangements from six observations of two simulated vaccination subjects, with missing-value and count limits visible — A seeded artificial vaccination study tests native table export. Row counts describe this example, not a biological finding.

Exact values within the native model

Counts use nullable signed integers; missing stays null and measured zero stays zero. Native counts must be nonnegative integral values no greater than 9,007,199,254,740,991 (2⁵³−1). Other counts refuse the export rather than being rounded. Parquet cannot recover precision lost before export. Boolean flags become true/false, while other retained AIRR annotations remain text, including invalid evidence.

Clonotype sample columns use keys such as sample_000001 with an explicit sample-name map. Clonotype and rearrangement units, effective QC and definition are recorded. Optional invariant-rule shares retain the full sample denominator and evaluable abundance; sequence-rule populations may overlap and are not established phenotypes or specificity assignments.

Choose pseudonymization separately

For a separate pseudonymized copy, choose Typed Parquet under De-identification export, make explicit free-text/HLA removal choices, then prepare and review before downloading. Public table footers and schemas carry sanitized provenance. The private identity map remains a separate download. Changing format clears the review.

Prepared Parquet pseudonymization review showing all subject correspondence, removal choices and separate public export and private map buttons — The local correspondence is private. This simulated review preserves subject/sample linkage without putting the private map in the table ZIP.

Parquet is a storage format, not a privacy guarantee. Sequences, stable clonotype IDs and rare measurements remain linkable. Read the de-identification scope and limits; unsupported nested provenance still refuses the AIRR option.

Keep evidence and its limits

The footer records the versioned schema and notice. The inventory checks recorded bytes; it does not certify biological validity, authorship or replay. Ordinary source fingerprints reflect loading, not a fresh check of files on disk. Keep original evidence for exact replay. Provider restrictions remain authoritative.

This slice maps native dataset tables only. Saved analysis tables, arbitrary nested results and Parquet import are outside its scope. Bounds and cancellation prevent partial successful exports. Reviewed automation, MCP, Python and R use the same explicit permissions and host downloads.

See the complete schema, bounds, reader examples and independent validation. The implementation follows the Apache Parquet format and retains native AIRR evidence. Independent PyArrow and DuckDB readers check typed values and provenance; this introduces no new estimator or scientific plot.