std::bad_alloc when running --gsmr2-beta flag

[juliaapolonio]

Hi! Thanks for the effort on this tool!

I am having some (around 10% of my pqtl data) bad allocation issues when running it with --gsmr2-beta flag. I tested in an exclusive node of 3TB RAM so I don't think I am facing memory issues. This is the final part of the log:

I didn't check all failed logs but it looks like this error happens when there is only 1 index SNP from Exposure. This creates a huge core dump (around 4GB of core file) and gives this general error on the log. When running without the flag, the task succeeds.

To replicate this issue,

• I am running with this container: quay.io/biocontainers/gcta:1.94.1--h9ee0642_0

• The sample files can be downloaded using this link: https://zenodo.org/api/records/14056241/files-archive

• The command I ran was:
  gcta64 --mbfile reference.txt --gsmr-file GORASP2_Q9H8Y8_OID31438_v1_Oncology_II.txt_format.input.txt outcome.txt --gsmr-direction 0 --gsmr-snp-min 1 --diff-freq 0.5 --gwas-thresh 5e-8 --clump-r2 0.05 --heidi-thresh 0.01 --effect-plot --gsmr2-beta --out "GORASP2_Q9H8Y8_OID31438_v1_Oncology_II.txt_format_finngen_sumstats"

Where the mbfile and gsmr-file options are sample sheets with name and path of the zenodo files.

As far as I understand, GSMR2 works by excluding the smallest-p IV iteratively and this is why it fails with 1 index SNP, so I suggest as an improvement to create a check on the number of index SNPs before proceeding to HEIDI analysis to avoid this massive core dump.

Thanks,
Julia