search

xinhe-lab / ctwas

package for the causal TWAS project
https://xinhe-lab.github.io/ctwas/
MIT License
32 stars 12 forks source link

Fully parameterize by-chromosome functionality in `ctwas_rss` | `impute_expr_z` maybe also need small tweak to accomodate #8

Open shugamoe opened 11 months ago

shugamoe commented 11 months ago

I've found it convenient when using cTWAS with sQTL models across multiple tissues (~20k models per tissue) to customize cTWAS to allow for by chromosome ctwas_rss parameter estimation/fine-mapping so that I'm not waiting more than 24 hours for a single tissue to finish running on one computer (even with single node parallelization), but instead waiting under that time for 22 chromosomes to finish independently instead.

Below is a link to some of the little hacks I made to allow for this functionality before the latest round of updates like pre-harmonization, separate parameter estimation/fine-mapping. (Some small messing around with the impute_expr_z output is required as well.) https://github.com/xinhe-lab/ctwas/compare/multigroup...shugamoe:ctwas:multigroup_dev

I will update this issue with the appropriate branch from my fork when I port these hacks over to the latest multigroup updates that have been made, but I think it would be nice for ctwas_rss to be parameterized to allow for at least param. estimation and finemapping by chromosome. I am aware of an example Wesley made to run ctwas for a single region, but I'm not a fan of also having to subset the .db model as well. Now that parameter estimation and fine-mapping can be separated, it might even make sense to do genome-wide parameter estimation, but by-chromosome fine-mapping/final PIP calculations (with each chromsome on a multicore node).

Ideally, it would be nice if when specifying chromosome in impute_expr_z (already possible), that the output for this is able to transfer smoothly to ctwas_rss so that a chromosome can be run singly in that function as well. I.e. modify ld_exprfs from impute_expr_z results if a chromosome is specified, so that it can integrate more smoothly into a chromosome parameterized ctwas_rss function.

shugamoe commented 10 months ago

https://github.com/shugamoe/ctwas/commit/766ed9ca6533d8a5f9314cc78516782198e719cc

This branch/commit is based off of the latest xinhe-lab/ctwas/multigroup and addresses the by-chromosome functionality (but not the tweak to ld_exprfs) as well as #7.