MR QC - how can strand issues for palindromic SNPs happen with modern genotyping?

ISG Workshop 2026 Day 7
1

In the first video about sensitivity analysis for Mendelian Randomization, it mentions that people often exclude palindromic SNPs because of the possibility of strand issues (but that sometimes things like allele frequencies can give you a clue whether that’s a concern or not).

I wanted to ask - are strand errors still a concern with modern genotyping technologies?

As far as I understand, on a genotyping chip, a SNP is found for genotyping based on the sequence of variants flanking it. So one probe might be ACGAGTTATAG and the other might be ACGAGATATAG

And if you’ve got sequencing data, it’s assembled by aligning reads to a reference, which also involve blocks of nucleotides and not just single SNPs.

In both cases, it seems like a strand error would not be possible unless:

  1. The sequence flanking the SNP was also palindromic, eg. ACGAGACTCGT (which, in reverse-complement, becomes ACGAGTCTCGT) - the probability of that would get lower the longer the sequences surrounding the SNP were.
  2. In the case of a genotyping chip, the chip company made a mistake when designing probes and made one that matched the wrong strand.
  3. Some other explanation I haven’t thought of.

Are strand errors actually still common enough with modern genotyping technologies that we have to worry about them? Or are palindromic SNPs mostly not a concern any more?

2

It’s not that the technology is producing an “error”, but rather that the results may not have been annotated properly. If you are combining results from a SNP-exposure association study and results from a SNP-outcome study you need to guarantee that the genotyping (whether sequencing or array based) has been performed on the same strand.

3

Thanks!

So if there was a mismatch, it would be on most SNPs (including non-palindromic) and it would stand out pretty clearly?

4

If you have one GWAS summary file where everything is on the forward strand and another where it is essentially random wrt strand- then yes it should stand out pretty clearly.