omicverse.micro.fetch_franzosa_ibd_2019

omicverse.micro.fetch_franzosa_ibd_2019#

omicverse.micro.fetch_franzosa_ibd_2019(data_dir, overwrite=False, microbe_count_scale=1000000.0)[source]#

Download + parse the Franzosa et al. 2019 paired IBD dataset.

Files are fetched from the Borenstein lab’s curated microbiome-metabolome-curated-data GitHub repository — three TSVs (genera.tsv, mtb.tsv, metadata.tsv) totalling about 30 MB. Once the files exist in data_dir the function is offline.

Parameters:

  • data_dir (str) – Absolute path the three TSVs are cached under. No $HOME fallback — you pick where it goes (recommended: a scratch directory).

  • overwrite (bool (default: False)) – Re-download even if the files already exist.

  • microbe_count_scale (float (default: 1000000.0)) – The Borenstein TSV delivers per-sample relative abundances. To make the tables look like familiar 16S count matrices (integer counts, range 10⁰–10⁵) we multiply by this scale and round — a pseudo-count-per-million by default. Pass 1.0 to keep proportions (most useful if you plan to CLR-transform immediately and don’t need integer counts). All downstream ov.micro APIs (filter_by_prevalence, paired_spearman, paired_cca, MMvec) work on either, but min_count filters expect counts ≥ 1.

Return type:

Tuple[AnnData, AnnData]

Returns:

  • (adata_microbe, adata_metabolite) — two AnnDatas sharing

  • obs_names (same 220 samples, same order). The microbe var

  • carries parsed GTDB 7-rank taxonomy (``domain / phylum / class /

  • order / family / genus / species`` + the raw GTDB string as

  • taxonomy). The metabolite var carries the cluster ID and,

  • where annotated, the HMDB name (name column; NaN for

  • unannotated clusters). Both obs frames carry the same cohort

  • metadata from metadata.tsv (Study.Group = CD / UC / Control,

  • Subject, Age, Fecal.Calprotectin, drug covariates).

Contents