Data And Results

The workspace separates source data, derived datasets, and run outputs.

Data Roots

Canonical data paths are resolved by genomics.workspace and genomics.core.data_registry.

Important environment variables:

Variable	Purpose
GENOMICS_DATA_ROOT	Override the default root for datasets
GENOMICS_RESULTS_ROOT	Override the default root for results

Registered Dataset IDs

Use dataset IDs in configs when possible instead of hard-coded absolute paths.

Common IDs include:

Dataset ID	Purpose
1kg_high_coverage	Canonical high-coverage 1000 Genomes dataset
legacy_top3_1kg_high_coverage	Historical top3 dataset, kept for provenance only
variant_transformer_superpopulation	Materialized variant transformer dataset
variant_transformer_superpopulation_32k	32k-window variant transformer dataset
variant_transformer_pigmentation_binary	Variant transformer pigmentation target dataset

Validate data availability with:

genomics audit-data --dataset-id 1kg_high_coverage --fail-on-missing

See Data Registry for the canonical dataset layout, registered dataset IDs, and bcftools_chain artifact checks.

Results

Runs generally write under:

results/<pipeline>/runs/<run_name>/

Run directories usually contain:

File or directory	Purpose
manifest.json	Run metadata and status
config.yaml	Copy of the config used
resolved_config.json	Expanded/validated config
models/	Checkpoints and model artifacts
plots/	Figures and diagnostics
reports/	Reports and metrics