# Data Garden plant protein dataset, release 2026.1

Functional and compositional measurements of plant proteins, produced by the
Data Garden, a program of Food System Innovations, under one protocol set.
10 starting materials (flours and protein concentrates) were each extracted by up
to three aqueous routes (isoelectric precipitation, micellar isolation,
membrane concentration), and the extracts were characterized under the same
protocol set. 6 commercial protein isolates were not extracted: each was
hydrated as supplied (route HYD) and characterized under the same protocols. Not every sample has every measurement: each gap is listed in
`annotations`, with its reason.

The methods are those of the Data Garden Methods Compendium, version
1.20, included as `methods/Data_Garden_Methods_Compendium_v1.20.pdf`.

## Anonymity

Every ingredient is published under an anonymous code (DG-0001) and every
extract under that code plus its route (DG-0001-IPI). Supplier and product
identities are never published. Commodity and form are given.

## Tables

All tables are in `tables/` as CSV and Parquet with the same columns; the
columns are described in `data_dictionary.csv`; `tables/metrics.csv` gives
each metric's assay, unit and definition.

| Table | Rows | One row is |
|---|---|---|
| ingredients | 16 | a starting material: species, common name, NCBI taxon id, form, nitrogen factor and protein content as supplied |
| samples | 36 | an extract (or hydrated sample) |
| values | 2991 | one replicate of one metric on one sample |
| summary | 1492 | n, mean, SD, median, min and max per sample and metric, computed from `values` (plus derived quantities) |
| yields | 30 | the protein mass balance of one extraction |
| conditions | 838 | a method setting a sample was run under (oil load, protein concentration, pH, speeds, temperatures) |
| caveats | 90 | a note on how a measurement differed from the current method |
| annotations | 19 | a measurement deliberately not made, and why |
| methods | 9 | a lab notebook template version behind the values |
| curves | 702 | an instrument run that has a series in one of the curves tables |
| curves_turbiscan_scans | 584600 | one height on one Turbiscan scan: backscatter against height and time |
| curves_turbiscan_profiles | 21000 | one height on the first and last Turbiscan scan: transmission |
| curves_turbiscan_kinetics | 2923 | one Turbiscan scan: the cumulative stability index and zone means |
| curves_texture | 43750 | one point of a texture analyzer force-time curve |
| curves_dsc | 78253 | one point of a DSC thermogram, per gram of sample |
| curves_dls | 35216 | one point of a DLS intensity-weighted size distribution |
| gels | 20 | a publication-grade SDS-PAGE gel, with its image under images/gels |
| gel_lanes | 223 | a lane: its sample, condition and size-range shares |
| gel_bands | 2283 | a measured band: size, abundance, detector confidence tier and, when confident, its protein |
| photos | 92 | a notebook photograph of a sample at a stage of its extraction or an assay, with its caption; the JPEG is under images/photos |
| color_reference | 18 | one patch of the color card in the reference photo: CIE L*a*b* as read by the lab's colorimeter, with the nominal value where one exists |
| photo_corrections | 55 | one photograph with its four tapes, corrected so the tapes and the bench look as they do in the key photo; the copy is under images/photos_corrected |
| raw_files | 408 | an instrument file: shipped under raw/ or withheld, and why |

`images/gels/` holds each gel's image cropped to the lanes in use, with lane axes given in `gel_lanes` in the image's pixels.

Bands in `gel_bands` were found by a detector trained on a curator's marks; `detection_tier` is its confidence. High and medium bands count toward lane shares and protein calls. A low-tier band is a minor component visible on the gel but excluded from shares, calls and any conclusion; it is kept so the record is complete. Bands outside the calibrated window carry no tier and no mass.

`images/photos/` holds the notebook photographs listed in `photos`, re-encoded with the camera's orientation applied and no other metadata.

`images/reference/` holds the color key: the four pieces of colored tape that appear in every sample photograph, photographed beside a DGK Color Tools Digital Kolor Kard 5 x 7 (DKK) under the lab's light with the same camera. The card's maker publishes no color values. `color_reference` gives each patch's CIE L*a*b* as read with the lab's FRU Precise Color Reader, model WR-10QC, on 2026-10-21, and the values are shipped as read, not corrected. The 18 percent gray reads L* 49.02 against 48.9 nominal, so the lightness scale holds; the neutral patches' a* and b* offsets from zero are given in `delta_a` and `delta_b`.

`images/photos_corrected/` holds color-corrected copies, for viewing, of the photographs that show the four tapes: each is corrected so its tapes match the key photo and its bench (or, on the black bench, its white label card) matches one common white, so photographs taken under different light look alike. `photo_corrections` gives each correction and how close the tapes came. They are not color measurements.

`raw/` holds the instruments' native files, one zip per instrument format, renamed to public codes; `raw_files` lists every file, shipped or withheld. A file is withheld only when it embeds identifiers or holds runs outside this release; its data is in the curves tables.

Method versions in `methods` and `values.method_version` are the lab
notebook template versions recorded when each notebook was created. The
templates were revised in place, so a notebook's version does not show the
earlier emulsion capacity and emulsion stability procedures used in the
first months of 2026. A value measured under an earlier procedure therefore
reads `earlier (see caveat)` in `values.method_version`, and its caveat
describes how the procedure differed from the current method; `conditions`
gives the settings each sample was actually run under.

Every table carries the public ingredient code (`ingredient`), or reaches it
in one step: the curve tables through `curves.run`, `gel_bands` through
`gels.gel`. Joining on `ingredient` to `ingredients` gives each value its
crop species, common name and NCBI Taxonomy id.

Replicates are technical replicates. Solubility is a pH series (pH 4 to 8),
one value per pH. A missing measurement appears in `annotations`, never as a
silent gap.

## Provenance

Each value traces to its instrument run or notebook entry in LIMBBS, the
Data Garden's laboratory system, through a crosswalk kept by the Data
Garden. `manifest.json` gives the code version that built this release, the
build time, row counts and a SHA-256 checksum for every file.

## Versions

Each release is fixed once published. A correction is a new release with a
new version number; public codes never change or get reused between
releases.

## Funding

Funded by Food System Innovations.

## License and citation

CC BY 4.0 (`LICENSE.txt`). Cite as given in `CITATION.cff`; the DOI of this release is 10.5281/zenodo.23548545 (https://doi.org/10.5281/zenodo.23548545).
