Datasets:

Modalities:
Text
Formats:
text
Size:
< 1K
Libraries:
Datasets
License:
umi_data / README.md
mikessh's picture
Sequences and metadata only: drop results/
23363ad
|
Raw
History Blame Contribute Delete
2.13 kB
---
license: gpl-3.0
tags:
- umi
- molecular-barcodes
- sequencing
- immunogenomics
pretty_name: UMI benchmark data for migec
---
# umi_data
Fixtures and derived results for [**migec**](https://github.com/antigenomics/migec) — UMI barcode
extraction, correction and consensus assembly.
This dataset holds **sequences and metadata, and nothing else**:
- `ci/` - small fixtures, cut from public data, that the test suite and the documentation examples
run against.
- `SOURCES.md` - where every file came from and the exact command that regenerates it.
Permitted file types: `.txt`, `.md`, `.tsv.gz`, `.json`, `.fastq.gz`, `.fa.gz`, `.sam`, `.bam`.
**No reports, figures, logs or pipeline output.** A derived results table is output, not data, even
when it is a TSV - those live in the code repository next to the script that made them.
## What is deliberately not here
- **Raw reads that are one command away.** `SRR1763769` is 248 MB on ENA and a `curl` recovers it
exactly; `SOURCES.md` carries the command instead of the bytes.
- **Simulated corpora.** Regenerable from a seed, and the seed is in `SOURCES.md`.
- **Anything under access control.** Nothing in this dataset came off a private cluster.
## `ci/SRR1763769_umi0.5pct.fq.gz`
2.12 M reads of an HIV-1 Primer ID library, checked out on a 9 nt Primer ID, then cut down to
**all of the reads of 0.5% of the barcodes** — 9,824 reads over 623 barcodes.
Never: Not a 0.5% sample of the reads. At 16 reads per molecule, sampling reads gives molecules seen
once each: the MIG size distribution is destroyed and every consensus is a single read, so the
fixture silently stops testing the thing it was built for. The barcodes are selected by hashing
(`splitmix64` of the packed barcode, kept when `hash % 10000 < 50`), which is unbiased, nested —
a smaller fixture is a subset of a larger one — and reproducible from the definition in
`include/migec/subsample.hpp`.
The fixture keeps **15.77 reads per barcode** against the full library's 16.05.
```bash
migec refine ci/SRR1763769_umi0.5pct.fq.gz -o ref/
migec assemble ref/CTRL.fq.gz -o cons/
```