Fortunately, at the moment, all instance of multiple-ref are in the similar format as shown above, I.e. there were at least one bp shared between snps.

This issue may come from the following setup in our vcf_qc module.

# when incorrect or missing REF allele is encountered: warn (w), no left normalization is done.
    bcftools norm -d exact -N --check-ref w -f ${reference_genome}  -Oz --threads ${numThreads} |\

The problem should disappear once we have a good target file that have only 1 ref

Two things need to be done:

• Fix vcf_qc to make this issue disappeared: For future user
  This should be simple enough based on following, test pending

-c, --check-ref e|w|x|s
what to do when incorrect or missing REF allele is encountered: exit (e), warn (w), exclude (x), or set/fix (s) bad sites. The w option can be combined with x and s. Note that s can swap alleles and will update genotypes (GT) and AC counts, but will not attempt to fix PL or other fields. Also note, and this cannot be stressed enough, that s will NOT fix strand issues in your VCF, do NOT use it for that purpose!!! (Instead see http://samtools.github.io/bcftools/howtos/plugin.af-dist.html and <http://samtools.github.io/bcftools/howtos/plugin.fixref.html>.)

• Create the correct sumstat reference: For us, cuz we don't want to spend couple days to redo all the genotype processing

After introducing stand alone TARGET file, this problem is fixed.