Documentation

Formats, APIs and the practical details of working inside a controlled-access environment.

Formats

Type	Format	Reference	Typical size
Alignments	CRAM 3.1	GRCh38 and T2T-CHM13	38–44 GB per 30× WGS
Per-sample variants	gVCF, bgzipped	matching build	1.4–2.2 GB
Joint callsets	VCF 4.3, sharded	matching build	0.2–2.1 TB per cohort
Structural variants	VCF 4.3 with BND	GRCh38	8–40 GB
Methylation	bedMethyl	GRCh38	2–6 GB

References

Both builds are maintained. Callsets state their reference in the header and in the STAC-style item metadata; we do not lift over silently.

Beacon v2

Open, unauthenticated, aggregate only. Answers whether an allele is present in a cohort and at what frequency bucket.

bash

curl 'https://beacon.usmirror.123ok.uk/api/g_variants?\
referenceName=17&start=43093000&referenceBases=C&alternateBases=T'

htsget

For approved users, htsget lets you fetch a genomic region from a CRAM without transferring the whole file — the difference between 40 GB and 8 MB.

bash

curl -H "Authorization: Bearer $HB_TOKEN" \
  'https://htsget.usmirror.123ok.uk/reads/HB0004821?\
referenceName=17&start=43044295&end=43125483'

htsget is the right tool for targeted work and the wrong tool for a genome-wide scan. If you are going to touch every region of every sample, run it in the enclave instead — the archive is already there and the transfer is zero.

Manifests

Approved bulk transfers are driven by a signed manifest containing URLs, sizes and checksums. Use a client that verifies checksums; at these volumes silent corruption is not theoretical.

Enclave basics

Log in through your federation, not a shared account
Projects get a workspace, a scratch volume and a core quota
The archive is at /archive, read-only
There is no internet — use the internal mirrors

Export review

Submit a directory. A reviewer checks it against your approval for individual-level data and for anything that could re-identify a participant. Median turnaround is four hours in working days.

Do not attempt to encode individual-level data into results files. Exports that appear designed to evade review are reported to your institution and to the committee.