indiv_report_i fails due to zero-variance scores on chromosome 22 test

[rinape]

Hey,
first, thank you for developing this pipeline!

I tried running GenoPred on the UK Biobank data and installed it via Conda following the instructions. To quickly test the pipeline, I restricted both the target genotypes and the GWAS summary statistics to chromosome 22 only.
I loaded the reference data for the methods ldpred2, quickprs, and sbayesrc using the Snakemake rules (the ldpred2 reference was downloaded from Google Drive). The pipeline ran successfully until the indiv_report_i step, where it failed because the variance of the score files was too low (all zeros).

I’m wondering:

• Is this failure expected when only a single chromosome is used, and would using multiple chromosomes or genome-wide data be required?
• Or is there another mistake I might have made before reaching this step?

Best,
Marina

[opain]

Dear Marina,

Thanks for reaching out.

Yeah, the indiv_report_i step will often fail when using only chromosome 22 as the estimated R2 of the PGS is too low to convert the PGS Z scores into absolute estimates of risk.

If you are using the pipeline for large target samples, such as UKB, I would strongly suggest turning off the indiv_report step, as otherwise it will create a report for each participant, which you probably don't want and will take a long time. To turn it off, just set the indiv_report column in the target_list file to F for the target sample: https://opain.github.io/GenoPred/pipeline_readme.html#target_list

Let me know if you have any other questions.

Best wishes,

Oliver

[rinape]

Hey,

thanks for the response!
I have another question regarding the "sbayesrc" method. I’m currently not getting any results for it

The *-TRANS.profiles files contain only zeros, while the other methods produce values.

For downloading the reference data, I used the get_sbayesrc_resources rule. However, after extraction, I had to manually move the files into the ancestry folder for them to be recognized. Could this be causing the issue, or is there something else I might be missing?
(We also set sampling and prevalence.. is this needed for this mehtod?)

Thanks again for your help!

Best,
Marina

[opain]

Hi Marina,

Sounds like you are making good progress.

I am not sure why the reference data would not be downloaded to the right place. I will check why this might be.

In the meantime, could you send me an example log file from the SBayesRC rule. There will be one in the logs folder, and in the same folder as the SBayesRC score file.

It would also be useful to see your config file.

Many thanks,

Ollie

[rinape]

Thanks for you help

The Configfile looks like this:

Specify output directory

outdir: results/output/

Location of this config file

config_file: ukb/config.yaml

Specify location of gwas_list file

gwas_list: ukb/gwas_list.txt

Specify location of target_list file

target_list: ukb/target_list.txt

Specify pgs_methods

pgs_methods: ['ldpre2','quickprs',‘sbayesrc‘]

ldpred2_ldref: resources/data/ldpred2
quickprs_ref: resources/data/quickprs
sbayesrc_ldref: resources/data/sbayesrc_ldref

Specify if you want test mode. Set to NA if you don't want test mode

testing: chr22

Specify number of cores for polygenic scoring methods (ptclump always uses 1)

cores_prep_pgs: 5

Specify number of cores for target scoring

cores_target_pgs: 5

Specify memory in megabytes for target scoring

mem_target_pgs: 2000

prep_pgs_sbayesrc_i-PGC044PRE.log
ref-PGC010.tidy.log
prep_pgs_sbayesrc_i-PGC010PRE.log
ref-PGC010.log

[opain]

Hi Marina,

Thank you for these files.

SBayesRC is not able to converge properly because you are only using chromosome 22 data. SBayesRC should work as expected if you are using genome-wide sumstats.

SBayesRC is different to other PGS methods as it requires GWAS sumstats for all variants in the LD reference. Step 2 of SBayesRC is sumstat imputation, which does not work since it only has SNPs on chromosome 22. This then causes issues downstream, leading to a SNP-based heritability of 0, and all SNPs having an effect of 0. In the future, I will try and make quick test case that will work for all methods.

Also, 'ldpre2' should be written as 'ldpred2', in the config file.

I can't figure out why the SBayesRC reference data would not have been downloaded to the correct directory. It should download the data into resources/data/sbayesrc_ref, which is the default location that the pipeline looks for the data.

Let me know if you have any other questions. I think you have seen it, but in case not, I provide guidance for running GenoPred on UKB-RAP here: https://opain.github.io/GenoPred/running_on_dnanexus.html

Best wishes,

Ollie

[rinape]

Hey,

Thanks for your response.

I was able to run it now and also obtained a score for sbayesRC. However, I couldn’t download the tools from https://cnsgenomics.com, as I wasn’t able to access the site. Do you think the following versions could be used instead?

https://gctbhub.cloud.edu.au/software/gctb/download/gctb_2.03beta_Linux.zip
https://gctbhub.cloud.edu.au/software/gctb/download/gctb_2.5.2_Linux.zip

Regarding the moving issue for the reference data, I encountered the following error message:
ukbEUR_HM3/block148.eigen.bin: Cannot change ownership to uid 1821486, gid 100: Invalid argument

As for running it with the DNAnexus tutorial, I don’t have experience with the Workbench and thought it would be faster this way :D

Best,
Marina