Answers are grounded in this paper and cite their sources.
A systematic, PRISMA-guided review led by our consortium screened 23,781 human microbiome papers and procured 2896 studies published from 2012 to 2025 that were then scored by reviewers into four Biological Host-Associated AI-Ready Metadata (BHAM) tiers. Only 5.9% of studies deposited sequence data with per-sample host disease status, age, and sex, and 63.1% shared no reusable sequence data, so that 85.5% of roughly 720,000 samples cannot be reused for cross-cohort or artificial intelligence (AI) analyses without contacting the authors. Sequence deposition improved modestly over time, whereas per-sample annotation did not. Journals, repositories, and researchers should require structured, machine-readable metadata at submission, supported by AI-assisted curation.
No takes yet. Share an insight, caveat, or question.
Degregori et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: