Add LOCO support for Plink format (Bimbam is in 0.97) about gemma

[pjotrp]

I am studying the GEMMA code to see how we add LOCO support.

[pcarbo]

Okay @pjotrp. It basically would involve an "outer loop" that would run relationship matrix creation + LMM association analysis on each chromosome. Note that I have some R code (see function run.gemma in that file) that does this in case it is useful for helping to implement LOCO within gemma.

[pjotrp]

The outer loop may be the easiest hack for now though it will be hard to make generic for all supported input formats. If you look at the code you'll see that GEMMA has an innerloop for every file type it supports. I somehow need to plug in K-1 and subset SNPs. Looks like that will be pretty intrusive to make that generic.

[pjotrp]

I have implemented LOCO for Bimbam format. I'll put in a PR after we agree on the coding style, see #62

[pjotrp]

Two more things need to be done:

1. Add LOCO to plink format
2. Run all chromosomes instead of just one

I think we need to focus on BIMBAM and Plink formats for core algorithms. Other formats should really format into one of these and eventually we should remove internal support. That way we can become more DRY.

wdyt?

[xiangzhou]

I agree.

[pcarbo]

@pjotrp Sounds like a good plan to me!

[pjotrp]

https://github.com/genetics-statistics/gemma-wrapper now runs all chromosomes and caches results too.

Leaving issue open until I add LOCO to Plink.

[pjotrp]

AnalyzeBimBam and AnalyzePlink contain duplicate logic - and they
differ now because the first has LOCO support. To add LOCO support to
PLINK we'll converge on logic and make the functions DRY.

The functions can be split into:

1. Initialization
2. Partially read genotype data (batch processing)
3. LMM on the submatrix
4. Collate the results
5. Cleanup

The only thing specific regarding the input file format is (2). The
way to solve is is to put that reader in a function that gets passed
in. I'll work on that in a new branch (for 0.98 release).

[flaviahodel]

Hello! Are there any news regarding LOCO for Plink format?

[pjotrp]

It will happen sometime this year. If you want to help out :). Meanwhile, convert plink to BIMBAM. It can be done with the plink tool.

[pjotrp]

It is part of gemma2/lib. Gemma2/lib converts plink and bimbam to a new Rqtl2/GEMMA2 format and that supports LOCO.