Running quantitative traits

emkoch / smilenfer

GNU General Public License v3.0

0 stars 0 forks source link

Running quantitative traits #1

Open A-Thijssen opened 1 month ago

A-Thijssen commented 1 month ago

Hi,

I want to run smilenfer for some quantitative traits. I have been able to reproduce the test trait (bc), but am wondering what I should change in the analysis pipeline to run quantitative traits. I have noticed that I should at least change some things in the test_run.yml file, but am not sure which parts to change and which not. Also, should I change anything else somewhere?

Thanks a lot! Anaïs

emkoch commented 1 month ago

Hi Anaïs,

There is actually nothing in the analysis pipeline specific to quantitative versus disease traits.

This part of the config file references trait types:

trait_files:
  BC: "clumped.genome_wide_ash.bc.max_r2.tsv"
trait_types:
  BC: "disease"
trait_type_abbrevs:
  disease: "Disease"

The entries here about trait types don't affecting anything except some plots.

If you are comfortable then following the pipeline in example_trait to process the GWAS summary statistics then nothing else in the config file would need to be altered aside from adding the new traits.

Thank you for trying out our software. Please let me know if that solves things or if there is anything else I can help with or improve in the code.

Evan

A-Thijssen commented 1 month ago

Hi Evan,

Thanks for your reply! I ran the software for the continuous trait and have a couple more questions:

Where can I find the p values that go with the figure 3 plot in the preprint? I can find the plot itself and the excel sheet that goes with it, but am unsure about what all the headers (ll_neut I2_stab ll_stab I1_dir ll_dir I1_full I2_full ll_full Ip_plei ll_plei I2_nplei nn_nplei ll_nplei) mean and how to transform the values to p-values.
In what direction is the directional selection tested? Is it both ways or only trait increasing? Should I change the sign for the beta's for a trait where I want to know if there is directional selection lowering the trait?

emkoch commented 1 month ago

Hi Anaïs,

The columns with ll_* in the title are log likelihoods. From these you can can obtain p-values for a likelihood ratio test where the neutral model is treated a the null hypothesis. For instance: 2*(ll_stab-ll_neut) has a Chi2 distribution with one degree of freedom. The script I used to generate figure 3 is here and you may use the plotting function. Let me know if you have any issues with it. Also please ignore the nplei columns for now as there are some issues currently with how this is calculated.

Directional selection is tested in both ways so you don't need to change the sign and re-run anything. If I1_dir is negative this means the fitted model has negative selection against trait-increasing alleles and if it is positive there is negative selection against trait-decreasing alleles.