GSL ERROR: function value is not finite in brent.c at line 202 errno 9

[peterdfields]

Hi,

I'm running into the following the error GSL ERROR: function value is not finite in brent.c at line 202 errno 9 when running gemma -bfile datafile -k datafile.cXX.txt -lmm 2 -maf 0.05 -o phenotype. The versions of gemma is 0.98.1 and it's running on a cent os 6 machine. When I run the same command with the -check flag I get the message FATAL ERROR: GEMMA caused a floating point error which suggests machine boundaries were reached. You can disable floating point tests with the -no-check switch (use at your own risk!). I tried to run the command with both the -no-check and no-fpe-check flags but the gemma analysis is still unable to finish and I see the same error. I tried subsetting the overall data to only include 100% genotyped individuals in case the error is being caused by something similar to issue #184 but the error still arises. Please let me know if any specific info would be useful in tracking down the cause of this error. Thank you for your time and advice.

[pjotrp]

Hmmm. If you send me your data files I can try and see what the problem is.

[peterdfields]

Hi pjotrp,

Is there an email address I could send a link to the file to?

[pjotrp]

pjotr public 673 at thebird dot nl

[pjotrp]

Looks like the number of phenotypes is very low.

[emc2rae]

Were you able to determine the cause of this error? I am running into it as well. I've narrowed it down to a single feature that seems to cause the error, and making tiny changes to the values of that feature makes it work.

EDIT: FWIW, this issue seemed to occur with features that have very small differences between their min and max values, and, so far, I've been able to get around it by scaling all my features to have a min of 0 and a max of 2.

[pjotrp]

Scaling is a good idea

[hrwang]

I am having the same error, and scaling is not helping. In my case, I have other evidences to suggest that the error is likely to be caused by a precision problem: there are several SNPs that are highly associated with the phenotype, p value maybe lower than 1e-60, maybe it reaches the limit of precision of the computer? My dataset also do not have many phenotypes, not sure it also affects. Could you take a look at my files if I send you the data file?

[pjotrp]

Yes, I can take a look. You can E-mail me the files if you want. See the footer of https://thebird.nl/

[hrwang]

Yes, I can take a look. You can E-mail me the files if you want. See the footer of https://thebird.nl/

This update fixed my issue. Now at these SNPs, I got nan, and the main program was not crashing.

[pjotrp]

Reopening to publish fix.

[pjotrp]

The fix is part of the 0.98.5 release I just put out!

[sql647]

I got a gsl error, too. I tried to use a parallel program to run gemma because I have a lot of phenotype. And now I'm in this follow error.gsl error：function value is not finite in brent.c at line 57 errno 9

[sql647]

I got a gsl error, too. I tried to use a parallel program to run gemma because I have a lot of phenotype. And now I'm in this follow error.gsl error：function value is not finite in brent.c at line 57 errno 9

And another errorgsl error：function value is not finite in brent.c at line 58 errno 9

[leegene1992]

Hi Pjotr,
It seems 0.98.5 release does not completely solved gsl error. It solved the problem when there are many SNPs that are highly associated with the phenotype, like p value is almost 0 or 0, but not with many SNPs that are highly associated with control.

[pjotrp]

Interesting. Do you have a data example you can share so I can look at providing a fix? Thanks. Also paste the error here.

[leegene1992]

Hi Pjotr,
Thanks for the quick response. Since my vcf is too large, I split it into different trunks. When running these trunks, some succeeded running, some not. Here I show you two cases (one succeeded, one failed) as example.

###the succeeded one
command:
gemma -bfile Chitili_Creek.split_5.txt.vcf -gk 1 -o relmatrix
gemma -bfile Chitili_Creek.split_5.txt.vcf -k output/relmatrix.cXX.txt -n 1 -lmm -o Chitili_Creek.split_5.txt.vcf
Logfile:
number of total individuals = 12
number of analyzed individuals = 12
number of covariates = 1
number of phenotypes = 1
number of total SNPs/var = 332207
number of analyzed SNPs = 263229
Start Eigen-Decomposition...
pve estimate =2.10768e-06
se(pve) =1.90547
**** WARNING: Brent did not converge 46%
**** WARNING: Brent did not converge= 76%
================================================== 100%
**** INFO: Done.

###the failed one
command:
gemma -bfile Chitili_Creek.split_4.txt.vcf -gk 1 -o relmatrix
gemma -bfile Chitili_Creek.split_4.txt.vcf -k output/relmatrix.cXX.txt -n 1 -lmm -o Chitili_Creek.split_4.txt.vcf
Logfile:
number of total individuals = 12
number of analyzed individuals = 12
number of covariates = 1
number of phenotypes = 1
number of total SNPs/var = 339566
number of analyzed SNPs = 268344
Start Eigen-Decomposition...
pve estimate =2.10194e-06
se(pve) =1.28465
GSL ERROR: function value is not finite in brent.c at line 57 errno 9

It seems GSL ERROR still exists. Here are the input files:

https://drive.google.com/drive/folders/14VzCub0hj5D-eyobGlildlcIITJbIOa6?usp=sharing

I would really appreciate if you can give me some help.
Best,
Jing

[pjotrp]

Wait, gemma does not take VCF files. I should give a warning on that.

[leegene1992]

Wait, gemma does not take VCF files. I should give a warning on that.

Sorry for the confusion. The Chitili_Creek.split_4.txt.vcf here is the prefix of plink file, but not VCF files.

[pjotrp]

Gotcha. I don't have the time to look into this right now, but if you leave the files I'll definitely take a look. Ping me in a week or so.

[pjotrp]

It may be a bit tough to fix. I can reproduce the error with the following stack trace. It is happening in the call to

gsl_root_fsolver_set((gsl_root_fsolver*)s_f, &F, lambda_l, lambda_h);

Thread 1 "gemma" received signal SIGINT, Interrupt.
0x00007ffff5e1eeca in raise ()
   from /gnu/store/fa6wj5bxkj5ll1d7292a70knmyl7a0cr-glibc-2.31/lib/libpthread.so.0
(gdb) bt
#0  0x00007ffff5e1eeca in raise ()
   from /gnu/store/fa6wj5bxkj5ll1d7292a70knmyl7a0cr-glibc-2.31/lib/libpthread.so.0
#1  0x00000000004245ca in gemma_gsl_error_handler (reason=<optimized out>, file=<optimized out>, 
    line=57, gsl_errno=<optimized out>) at src/gemma.cpp:75
#2  0x00007ffff7e5aa4d in brent_init ()
   from /gnu/store/hkzqndw8jik4vf48yf4kii8cz9y9lps1-profile/lib/libgsl.so.25
#3  0x0000000000472338 in CalcLambda (func_name=func_name@entry=82 'R', params=..., 
    l_min=l_min@entry=1.0000000000000001e-05, l_max=100000, n_region=<optimized out>, 
    lambda=@0x7fffffffcc98: nan(0x8000000000000), logf=@0x7fffffffcc70: nan(0x8000000000000))
    at src/lmm.cpp:2027
#4  0x00000000004762b8 in LMM::AnalyzePlink (this=this@entry=0x7fffffffd170, U=U@entry=0x4fbb20, 
    eval=eval@entry=0x4fbb60, UtW=UtW@entry=0x4fbb90, Uty=Uty@entry=0x7fffffffd110, 
    W=W@entry=0x4fd230, y=<optimized out>, gwasnps=...) at src/lmm.cpp:1863
#5  0x0000000000435a05 in GEMMA::BatchRun (this=this@entry=0x7fffffffe6f0, cPar=...)
    at src/gemma.cpp:2798
#6  0x000000000047c534 in main (argc=10, argv=0x7fffffffe858) at src/main.cpp:86

[pjotrp]

I made a fix. Try the static binary at https://biogems.info/download/gemma-0.98.6-pre1.gz

Note that 12 individuals is really not enough power to run GWA. Hits can only be suggestive.

[leegene1992]

Dear Pjotr, thanks a lot for the debugging. It works perfectly now. Cheers.

[sql647]

Hi, I updated gemma's version and the error stopped. I'm currently using gemma-0.98.6-pre1, and you can get the download link for this updated version from github. 2023-12-12 04:01:11"Jeff Groh" ***@***.***>写道： I'm experiencing the same error message. In my case I expect that some SNPs are perfectly associated with the phenotype, and I'm using a maf filter of 0.2. I'm not sure how to troubleshoot this issue. Any suggestions? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[jasongallant]

I'm having a similar issue to the previous posters. The download link for gemma-0.98.6-pre1.gz no longer works-- is there somewhere else this can be obtained from?

[jgroh]

I am also experiencing this issue using 0.98.5, and can't access the download link for the newer version. Is there a known fix for this issue?

[vivekruhelaIIITD]

Hello,

I am also getting the same error: GSL ERROR: function value is not finite in brent.c at line 57 errno 9.

The download link for GEMMA verrsion 0.98.6 (https://biogems.info/download/gemma-0.98.6-pre1.gz) is not working. Can you please share the link again.

Thanks.