search

jianyangqt / gcta

GCTA software
GNU General Public License v3.0
73 stars 23 forks source link

GCTB Issue #80

Open eromic96 opened 1 month ago

eromic96 commented 1 month ago

Hey,

I tried to run SBayesR today with a prepared summary statistics file and this is the error I get:

“Error: Residual variance is negative (-0.452340). This may indicate that effect sizes are "blowing up" likely due to a convergence problem. If SigmaSq variable is increasing with MCMC iterations, then this further indicates MCMC may not converge.”

It outputs results for Chromosome 13, 18, 20, 21, and 22, and give that error message for all others (with different number for the variance).

What do you think could be wrong and is causing this error message?

Thank you in advance!

jianzeng commented 1 month ago

This error message flags a convergence issue in most chromosomes. The failure to converge may be due to LD differences between GWAS and LD reference samples or heterogeneity in the GWAS summary statistics, such as unequal sample size across SNPs.

  1. Make sure you use a LD reference that is genetically as close as to the GWAS sample.
  2. Check the per-SNP sample size distribution and remove the outliers.
  3. You may want to try SBayesRC (can work without annotations) which uses a low-rank model to improve robustness. Tutorial: https://cnsgenomics.com/software/gctb/#SBayesRCTutorial

Best regards, JZ

From: eromic96 @.> Date: Friday, 24 May 2024 at 8:56 pm To: jianyangqt/gcta @.> Cc: Subscribed @.***> Subject: [jianyangqt/gcta] GCTB Issue (Issue #80)

Hey,

I tried to run SBayesR today with a prepared summary statistics file and this is the error I get:

“Error: Residual variance is negative (-0.452340). This may indicate that effect sizes are "blowing up" likely due to a convergence problem. If SigmaSq variable is increasing with MCMC iterations, then this further indicates MCMC may not converge.”

It outputs results for Chromosome 13, 18, 20, 21, and 22, and give that error message for all others (with different number for the variance).

What do you think could be wrong and is causing this error message?

Thank you in advance!

— Reply to this email directly, view it on GitHubhttps://github.com/jianyangqt/gcta/issues/80, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ABJ35RSV6Z2J75XVOYZ2BYDZD4MHHAVCNFSM6AAAAABIHLHFIGVHI2DSMVQWIX3LMV43ASLTON2WKOZSGMYTKMJQGM4DKNI. You are receiving this because you are subscribed to this thread.Message ID: @.***>