tooluniverse-dataset-discovery

Start by defining the required study design, variables, and population, then search in layers: cross-repository indexes first, domain-specific repositories like GEO or NHANES second, and literature methods sections third. Literature searches often reveal niche datasets not listed in any repository.

It covers health surveys like NHANES and CDC, genomics repositories like GEO, SRA, ENA, and ArrayExpress, proteomics via PRIDE and MassIVE, metabolomics via MetaboLights, and clinical trial data via ClinicalTrials.gov, plus 30+ other scientific repositories.

No. Cross-sectional data measures individuals once and cannot answer questions about changes over time. The skill explicitly flags this mismatch and recommends true longitudinal cohorts that follow the same individuals across multiple waves.

Use pd.read_sas with format="xport" for SAS transport files and pd.read_stata for Stata .dta files. The skill provides a data loading cookbook covering CSV, Excel, Parquet, JSON, fixed-width, and REST API pagination patterns.

Some datasets require registration, collaboration agreements, or are fully restricted, which can take days to months or be impossible to obtain. The skill distinguishes these access tiers and reports honestly when no accessible dataset matches the required study design.