PLINK conversion

vcztools view-plink transcodes a VCZ store into a PLINK 1 binary fileset (.bed/.bim/.fam). The format semantics are described in the PLINK 1 file format page; this page walks through the command itself.

The examples run against the example dataset data/sample.vcz.zip (see Reading VCZ).

Basic conversion

-o gives the output stem. The dataset contains multi-allelic sites, which PLINK cannot represent, so -M 2 is required to skip them (mirroring plink2 --vcf X --max-alleles 2 --make-bed):

!vcztools view-plink data/sample.vcz.zip -M 2 -o sample

This writes three files:

!ls sample.bed sample.bim sample.fam

sample.bed  sample.bim	sample.fam

The .bed holds the packed genotypes (binary). The .fam lists one line per sample and the .bim one line per variant:

!cat sample.fam

NA00001	NA00001	0	0	0	-9
NA00002	NA00002	0	0	0	-9
NA00003	NA00003	0	0	0	-9

!cat sample.bim

19	.	0	111	C	A
19	.	0	112	G	A
20	rs6054257	0	14370	A	G
20	.	0	17330	A	T
20	.	0	1230237	.	T
20	.	0	1235237	.	T

Subsetting by sample

Pass -s (or -S FILE) to convert only a subset of samples. Here we keep two of the three:

!vcztools view-plink data/sample.vcz.zip -M 2 -s NA00001,NA00003 -o subset

The .fam now lists just those samples:

!cat subset.fam

NA00001	NA00001	0	0	0	-9
NA00003	NA00003	0	0	0	-9

Only the selected samples are read from the store, so pulling a handful of samples out of a very wide cohort is cheap — the full cohort is never materialised.

Filtering over the sample subset

-i/-e expression filters that reference sample-derived INFO fields (AC, AN, AF, NS) are evaluated over the selected samples. With a -s/-S subset, those values are recomputed for that subset rather than read from the file’s stored (full-cohort) INFO. So view-plink -s cohort.txt -i 'AC>0 & AC<AN' keeps the variants that vary in your cohort (both alleles present, 0 < AC < AN), dropping sites whose ALT alleles are carried only by excluded samples — even where the source file’s stored INFO/AC is non-zero.

This is deliberately different from vcztools view and vcztools query, which follow bcftools and evaluate -i/-e against the original record’s stored INFO (so the same -i 'AC>0 & AC<AN' keeps a site on its stored full-cohort count). The rationale: a PLINK fileset is the input to an association analysis run on the exported cohort, so filters should reflect that cohort, not the source dataset. It is also the cheaper path on large datasets — only the selected samples are read.

Note this applies to expression filters. The allele-based filters (-m/-M/-v/-V) remain record-level.

Sidecars

The .bim and .fam sidecars are written by default; pass --no-bim or --no-fam to suppress them.

See also

• The PLINK 1 file format page — format details.

• Format conversion — the equivalent Python writer API.

• The view-plink flag reference.