inst/DATA_PROVENANCE.md

In linf: L-Infinity Normalization and Dominant Community State Types

Bundled data provenance and upstream terms

The package code is distributed under the package's MIT license. The bundled microbiome data are derived from the upstream resources described below. The package license does not replace or narrow applicable upstream attribution terms.

VALENCIA-derived objects

The valencia2k, valencia_linf_hypercube_1k, valencia13k_dcst_depth2_merged, and valencia13k_dcst_depth3_merged objects are derived from the VALENCIA training data:

• Source repository: https://github.com/ravel-lab/VALENCIA
• The primary article states that the training dataset is available in that repository and applies the Creative Commons CC0 public-domain dedication to data made available with the article unless otherwise stated.
• Upstream software repository license: MIT, copyright 2023 ravel-lab
• Primary reference: France et al. (2020), "VALENCIA: a nearest centroid classification method for vaginal microbial communities based on composition," https://doi.org/10.1186/s40168-020-00934-6

The source-preparation scripts in data-raw/ record the deterministic subsampling and dCST construction parameters. Original sample identifiers are not retained in the merged 13k assignment objects.

American Gut Project demonstration object

The agp_gut object is derived from public American Gut Project 16S records and deidentified metadata:

• ENA/BioProject accession: PRJEB11419 (https://www.ebi.ac.uk/ena/browser/view/PRJEB11419)
• Primary reference: McDonald et al. (2018), "American Gut: an Open Platform for Citizen Science Microbiome Research," https://doi.org/10.1128/mSystems.00031-18
• The primary article is distributed under CC BY 4.0. It states that the project protocol permits public deposition of non-identifying data and that sequence data and deidentified participant responses are placed in the public domain without access restrictions.
• Processing source: PRIME SILVA species-level outputs dated 2026-03-24.

The full analysis first selected at most 5,000 samples with library size at least 1,000 using random seed 42. data-raw/create_agp_gut_subset.py then creates the 766-sample package object using an explicit deterministic rule:

1. include every sample assigned to Prevotella_7, Pasteurellaceae, Akkermansia, or Staphylococcus;
2. exclude depth-1 Eukaryota and Unassigned labels; and
3. fill the remaining slots by a seed-42 simple random sample from the eligible background.

Phenotype fields are joined only after membership is fixed and cannot influence selection. Exact run-ID membership, selection reasons, dCST labels, and derived phenotype fields are retained in inst/extdata/agp_gut_meta.csv. The script also reconstructs the bundled count matrix from the full abundance table and retained 5,000-sample dCST assignments. data-raw/build_agp_gut.R then constructs the package object.

Because inclusion probabilities differ by dCST, agp_gut is suitable for demonstrating data alignment, filtering, normalization, and dCST construction only. Its phenotype frequencies, effect sizes, and p-values must not be interpreted as population estimates.

linf documentation built on Aug. 5, 2026, 9:08 a.m.