Conditional analysis with multiple variants

[Nobuyuki-Enzan]

Hi regenie team,

I am trying to run conditional analysis with 3 SNPs (SNP_A, SNP_B, SNP_C).
SNP_A is the lead variant in the unconditional analysis.
SNP_B is the lead variant when I ran regenie conditioning on SNP_A.
SNP_C is the lead variant when I ran regenie conditioning on SNP_A and SNP_B.
So SNP A, B, and C are independent variants.

However, when I ran regenie conditioning on SNP_A, SNP_B and SNP_C, regenie throwed the error
Please see the regenie.log.

block [143/165] : ERROR: Throw location unknown (consider using BOOST_THROW_EXCEPTION)
Dynamic exception type: boost::wrapexcept<std::domain_error>
std::exception::what: Error in function boost::math::cdf(const chi_squared_distribution<double>&, double): Chi Square parameter was -nan, but must be > 0 !

I tried two approaches.

1. Specified the conditional variants only at the step 2 stage and use the unconditional results from step 1
2. Specified the conditional variants at both step 1 and step 2
   but to no avail.

When I ran regenie condioning on only SNP_C, it completed successfully.
So SNP_C alone did not cause this error.

Next I ran conditional analysis after excluding variants with r2 > 0.8 with SNP A or B or C.
Regenie still throwed the same error at the same block (143/165).

Let's say SNP_D is the root cause of this error.
From what I observed, SNP_D is not strongly correlated with SNP_A or B or C (r2 < 0.8).
I suspect there is a strong multicolinearity between SNP_D and COMBINATION of SNP A B C.

If I'm correct, how can I identiy these variants (SNP_D)?
Or Is there any reason for this error?
Is there any way to skip these variants (SNP_D)?

I want to skip SNP_D and complete the conditional analysis with SNP A, B, and C.

regenie.log

Thank you,
Nob

[joellembatchou]

Hi Nob,

From the log, the issue is not with the variants specified in the conditional analysis as it is running fine across many of the blocks. There seems to be a variant in block 143 causing the error. You could get the list of variants in the sumstats file (ie which were tested successfully) and pass these to --exclude in a new regenie call then decrease the block size to narrow down which variant is causing the bug. We will work on addressing this error in the next release.

Cheers,
Joelle