18. Datasets, registries, reference resources

Benchmark and training data - computage/computage_bench on Hugging Face - 10,404 benchmark samples / 65 studies and 7,419 training samples / 46 studies, parquet + TSV metadata. - lucascamillomd/pyaging-data on Hugging Face, clock weights, reference vectors, Ensembl gene annotation, example datasets per modality. - GEO series listed in ComputAge BenchBSB_table.csv. - GSE54848 - OSKM reprogramming timecourse used in the pyaging figure. - NHANES III and NHANES IV - shipped as .rda in BioAge with a codebook. - UK Biobank, Framingham Heart Study, Generation Scotland, MESA, Health and Retirement Study, Dunedin Study, Tabula Muris Senis, GTEx.

Gene and biomarker resources - Open Genes (2,402 genes), GenAge, AnAge (the mammalian clock lifespan lookup), CellAge, CSGene, AgeFactDB, Aging Atlas, GO aging terms, Digital Ageing Atlas, AGEMAP.

Reference matrices - sesame_450k_median.csv - NA-fill medians. - 450K_reinius_12_reference.csv, EPIC_salas_18_reference.csv - deconvolution signatures. - biolearn_averages_450k.csv - alternative fill vector. - DunedinPACE_Gold_Means.csv - the 20,000-probe gold standard.

Model registries - pyaging clock_metadata.json (177 clocks at e7267e9) + evidence_ledger.jsonl (930 kB of per-field provenance) + controlled_vocabulary.json + validate_metadata.py. - tAge clocks_metadata.csv (60 models) + Zenodo record 18763485. - ComputAge model_definitions + library.yaml (GEO dataset parsers).