Low SNP imputation concordance - except chromosome 1

Question

[muraved]

Hi,

I'm trying to do SNP imputation using IMPUTE2. My genotype data (bim/bed/fam) is hg18, with the usual quality control, mostly European ancestry, reference panel is hg19.

My workflow:

1. Determine flipped and ambiguous SNPs relative to hg18, using snpflip (https://github.com/biocore-ntnu/snpflip ).
2. Use PLINK to create map/ped files, with flipped SNPs flipped and ambiguous removed.
3. Computed liftover to hg19 using liftOverPlink.py (https://github.com/sritchie73/liftOverPlink ).

4. Using gtools to transform ped/map to gen/sample. Final SNP counts per chromosome in the .gen file:

   63234   1
   64938   2
   53168   3
   49266   4
   49529   5
   49570   6
   41084   7
   42120   8
   35963   9
   42686   10
   39260   11
   37593   12
   30342   13
   24798   14
   22896   15
   23652   16
   18379   17
   23539   18
   10569   19
   20286   20
   11094   21
   10292   22
   

5. Using IMPUTE2 with hg19 reference panel (https://mathgen.stats.ox.ac.uk/impute/1000GP_Phase3.html ), chunk size 5000000, -Ne 20000 and -filt_rules_l 'EUR==0'.

I'm parsing the top-right entry in the concordance table (see https://mathgen.stats.ox.ac.uk/impute/impute_v2.html#concordance_tables ), using

cat $summaryfile | grep -A1 "Concordance" | grep -v "Concordance" | tr -s ' ' | cut -d ' ' -f 8,9 | grep -v -- -- | tr -d '\n'; echo

When plotting these values by chromosome, it turns out that chr1 has good concordance as expected (~95%), whereas all others are pretty bad:

The same happens when not filtering for EUR.

I'm at a loss here, any idea what could be causing this?