Oxford format with PLINK2

[Gad Abraham]

Hi Chris,

I've been using PLINK2 to process imputed genotypes in Oxford format
(.bgen), and it's amazingly fast and useful, thanks for all the hard
work!

A couple of questions:

- Is it possible to export gzipped dosages (.gen.gz) natively, i.e.,
without writing an intermediate .gen file and then gzipping it?

- Can PLINK2 compute the impute INFO statistic
(http://www.well.ox.ac.uk/~gav/qctool/#documentation, under 'Info')?

Thanks again,
Gad

[Christopher Chang]

1. While it would be straightforward to add this to PLINK2, I suspect it would create more problems than it solves.

The issue is, the .gen format includes separate P(hom ref), P(het), and P(hom alt) probabilities, whereas PLINK2 only keeps track of a single "alt dosage" value (since the other values take up quite a bit of space, while having minimal relevance to PLINK2's analysis functions).  I.e. {P(hom ref)=0.2, P(het)=0.52, P(hom alt)=0.28} and {P(hom ref)=0, P(het)=0.92, P(hom alt)=0.08} are both imported as "alt dosage = 1.08", and if you then ask PLINK2 to re-export this information, it'll always be represented as {P(hom ref)=0, P(het)=0.92, P(hom alt)=0.08}.  This "lossy compression" makes PLINK2 a poor choice for general-purpose .gen data management.

What I can do is create a mini-qctool which uses PLINK2's multithreaded .bgen loader, and can directly export a .gen.gz without throwing away any data.

2. The impute INFO statistic requires all three probabilities, so PLINK2 cannot compute it (instead, it will have an option for computing the similar MaCH r^2 statistic).  However, this would also be straightforward to support in a mini-qctool.

[Gad Abraham]

Right, I see the problem now.

How much work would it be to make a tool that can read .bgen and write
.gen/.gen.gz but also supports --extract and --keep? That would be
super useful for people using imputed data as existing tools are very
slow.

Gad

> --
> You received this message because you are subscribed to the Google Groups
> "plink2-users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to plink2-users...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Christopher Chang]

I can probably have that ready by next week, if only .gen{.gz} export, --extract, --keep, and INFO computation are needed.

On Tuesday, May 9, 2017 at 9:36:20 AM UTC-7, Gad Abraham wrote:

Right, I see the problem now.

How much work would it be to make a tool that can read .bgen and write
.gen/.gen.gz but also supports --extract and --keep? That would be
super useful for people using imputed data as existing tools are very
slow.

Gad

[Gad Abraham]

That would be really great, happy to test it out when you're ready.

Gad

[Gad Abraham]

One more question, when using --dosage together with --score, does
PLINK internally hard-threshold the dosages before computing the score
or does it treat it as a proper dosage?

Thanks
Gad

[Christopher Chang]

PLINK 1.x --score does actually work properly with --dosage (no thresholding).

On Thursday, May 11, 2017 at 2:51:51 AM UTC-7, Gad Abraham wrote:

One more question, when using --dosage together with --score, does
PLINK internally hard-threshold the dosages before computing the score
or does it treat it as a proper dosage?

Thanks
Gad

> That would be really great, happy to test it out when you're ready.
>
> Gad
>

[Gad Abraham]

Great!

And to confirm, the --bgen/--sample interface does do hard
thresholding internally when calling --score?

[Christopher Chang]

With v1.9, and not with v2.0.

On Thursday, May 11, 2017 at 9:14:32 AM UTC-7, Gad Abraham wrote:

Great!

And to confirm, the --bgen/--sample interface does do hard
thresholding internally when calling --score?

[Gad Abraham]

Could you clarify something:

I'm using PLINK2 2017-05-26 to do profile scoring on bgen v1.1 files.
As far as I understand, the above ambiguity in the genotype
probabilities shouldn't affect scoring since the expected dosage is
identical.

In the .sscore output file, what is the column NMISS_ALLELE_CT? I'm
getting values lower then the number of alleles (2x number of SNPs),
but there shouldn't be missing alleles, unless PLINK2 is interpreting
some dosages as missing?

[Christopher Chang]

You're interpreting NMISS_ALLELE_CT correctly.  "0 0 0" entries in the .bgen file are interpreted as missing dosages, and chrY variants only contribute one allele per SNP for males and zero for females; are either of these present in your data?

[Katherine Fawcett]

Hi Chris,

I was just wondering whether you wrote a mini-qctool for converting from bgen to gen.gz format (see e-mails below)?  If so, how do I use it (I'm new to PLINK...)?

Many thanks

Kath

[Christopher Chang]

Didn't end up getting around to it back then, sorry.  But if qctool2, etc. still can't handle this, I should be able to do it soon.