IBS multi-omics study — Supplementary Data S1 (derived summary tables)

This archive contains de-identified derived tables, not raw sequencing data.
Data provenance: Cell 2020 cohort (ENA PRJEB37924; processed matrices at
Mendeley Data doi:10.17632/29n2z5r5ph.3, CC BY 4.0) and Jacobs 2023 cohort
(PRJNA812699). Taxonomy/function abundances were obtained from GMrepo.

Files
-----
p0_cohort.csv
  Per-layer three-class (IBS-D/IBS-C/Healthy) OOF performance: n samples,
  n subjects, n features, pooled and fold-mean balanced accuracy, macro-F1,
  subject-cluster bootstrap 95% CI, permutation p and number of permutations,
  pooled 3x3 confusion matrix, null mean/SD. Permutation p floor is 1/(B+1).
p0_severity.csv
  IBS-SSS severity task (genus layer): OOF balanced accuracy, CI, permutation p.
p0_cohort_oof.csv / p0_severity_oof.csv
  Row-level OOF predictions used for the uncertainty analyses; subject IDs are
  the de-identified IDs used in the source matrices.
p0_combined_filtered.csv
  NMR+genus combined model under the same prevalence filter (87 features,
  n=424/67): OOF R2, rho, alpha, and Delta R2 versus NMR alone.
p0_external_baselines.csv
  Jacobs 2023 external-validation baselines (demographics, random 11 features,
  full 16 features, etc.).
threshold_sweep.csv
  Source-domain closed-loop stability-selection gate sweep used for Figure 11.
biomarkers.csv
  Spearman associations between NMR metabolites and Bristol score.
sample_map.csv
  De-identified sample-to-subject/cohort/Bristol mapping (464 labeled samples,
  75 subjects).
results_bristol_*/
  Per-layer Bristol regression OOF predictions (predictions.csv) and
  cross-validation summaries (cv_results.csv).

Counts used in the manuscript
-----------------------------
464 clinically labeled samples / 75 subjects (IBS-D 169, IBS-C 142, Healthy 153)
Genus/species 457 (Bristol 441/67)
KO/KEGG 459 (Bristol 437/67)
NMR 445 (Bristol 424/67)
Bile acids 447 (Bristol 426/67)
Tryptophan 274 (Bristol 259/67)
NMR+genus 440 (Bristol 424/67)
