# Filtering imputed GWAS SNPs based on a MAF difference of 10%

There are many posts on the web regarding QC steps pre and post-imputation.

Does applying below (new?) 10% MAF difference rule make sense, pitfalls?

Here is the process:

1. Get MAF for imputed set, using SNPTEST with flag -summary_stats_only
2. Convert imputed set to hard-calls using gtools with flag --threshold 0.9
3. If the MAF from step 1 and step 2 differs more than 10% than exclude the variant.

• GWAS is 50K vs 50K case control samples.
• This step is applied after info > 0.4 filter.

I cannot think of any principled rationale for choosing this filtering strategy.

However, I am going to take a guess that this filtering strategy is supposed to filter out SNPs for which imputation did not work well? In that case the appropriate statistic to filter on is the INFO score as described here.

You might consider picking a higher threshold than 0.4 if you feel that you aren't filtering out enough variants.