search

privefl / bigsnpr

R package for the analysis of massive SNP arrays.
https://privefl.github.io/bigsnpr/
186 stars 44 forks source link

LDpred2, snp_ldpred2_auto outputs NAs #423

Closed lingjoyo closed 1 year ago

lingjoyo commented 1 year ago

Hallo Florian

I'm testing this guideline for calculating PRS using LDpred2 https://choishingwan.github.io/PRS-Tutorial/ldpred/ The inputs are example data provided by the site. The package, bigsnpr, is version 1.12.2.

When I tried to get adjusted beta from the auto model using following codes, the outputs are all NAs. multi_auto <- snp_ldpred2_auto( corr, df_beta, h2_init = h2_est, vec_p_init = seq_log(1e-4, 1.0, length.out=20), ncores = NCORES, sparse = T,verbose = T ) Do you know the reason? How should I solve this problem?

lingjoyo commented 1 year ago

There is another problem with calculating heritability (h2) using the function snp_ldsc. I used GWAS of coronary artery disease (CAD) of which heritability was estimated 40%-50% in some paper. But the result from snp_ldsc is only 9%. Do you think it's normal?

privefl commented 1 year ago

  1. Please use the official up-to-date tutorial: https://privefl.github.io/bigsnpr/articles/LDpred2.html.

  2. Missing values are due to divergence issues (as stated in the documentation). Possible solutions at https://github.com/privefl/bigsnpr/issues/416#issuecomment-1487155785.

  3. Please open another separated issue for the LDSc heritability estimation.

privefl commented 1 year ago

Any update on this?