6
$\begingroup$

Hi I'm a first time user of IADMIX. I tested on one known Finnish sample from the 1000Genome project using the softwares provided frequency table hapmap3.8populations.hg19.freqs and the prediction is correct.

I'm also want to do the same using population frequencies from the gnomAD database but my prediction is wrong - comes out to be African.

My steps are: Create my own population frequency table using the gnomad database from http://gnomad.broadinstitute.org/downloads For each VCF file i do the following

bcftools view -Ou -I -q 0.05  -v snps -k -m2 -M2 -f PASS gnomad.genomes.r2.0.1.sites.22.vcf.gz | bcftools query -f '%CHROM\t%POS\t%ID\t%REF\t%ALT\t%INFO/AF_nfe\t%INFO/AF_afr\t%INFO/AF_amr\t%INFO/AF_eas\t%INFO/AF_asj\t%INFO/AF_fin\t%INFO/AF_oth\n'  -o iadmix-22.freq.bcf

which gives me tabulated data like this containing only "known" variants with RSID != "." , min threshold for total AF >= 0.05 , only bi-allelic SNPs that also have FILTER==PASS

7 16692 rs370739206 G C 1 . . . 1 . 1 7 16719 rs151261124 G A 1 1 1 . 1 . 1 7 20460 rs1466291194 T C 0.0381589 0.0630435 0.191919 0.0357143 0.0333333 0.144909 0.121212 7 20471 rs370790239 C T 0.631696 0.716216 0.766667 0.666667 . 0.661765 0.625

Then I have a small R script to - exclude sites that have any population groups AF values == "." - include sites that has mean of all AFs for the snps must be >= 0.05 (to include only common variants)

Then I tried running IADMIX using this frequency table on a know finnish 1KG sample, but the prediction is incorrect

python runancestry.py --freq MY.FREQUENCY.FILE --plink SAMPLE_file --out testoutput --path full/path/

snps in file 455423 pops 7 AF_NFE:0.167 AF_AFR:0.189 AF_AMR:0.133 AF_EAS:0.188 AF_ASJ:0.238 AF_FIN:0.037 AF_OTH:0.047 initial sol 1082621.766160 RUNNING THE L-BFGS-B CODE

    * * *

Machine precision = 2.220446e-16 ... ...

final maxval -1044248.568804 ADMIX_PROP AF_NFE:0.0000 AF_AFR:0.5822 AF_AMR:0.1137 AF_EAS:0.1797 AF_ASJ:0.1244 AF_FIN:0.0000 AF_OTH:0.0000 AF_AFR:0.5822 AF_AMR:0.1137 AF_EAS:0.1797 AF_ASJ:0.1244 FINAL_NZ_PROPS

Any one know whats happening here?

For reference the output using the frequency table provided by IADMIX is as follows

snps in file 106250 pops 8 YRI:0.015 CHB:0.003 CHD:0.275 TSI:0.207 MKK:0.101 LWK:0.169 CEU:0.194 JPT:0.037 initial sol 76610.617803 RUNNING THE L-BFGS-B CODE

    * * *

Machine precision = 2.220446e-16 final maxval -72537.379103 ADMIX_PROP YRI:0.0000 CHB:0.0000 CHD:0.0000 TSI:0.1031 MKK:0.0000 LWK:0.0000 CEU:0.8190 JPT:0.0779 TSI:0.1031 CEU:0.8190 JPT:0.0779 FINAL_NZ_PROPS

so the CEU proportion is 82% , acceptable for this finnish sample, right?

$\endgroup$
1
  • $\begingroup$ I’m voting to close this question because it seems to lack important detail. There's mention of a custom R script (not included) that is used for filtering SNP frequencies, and it's not clear where the rationale behind doing this filtering comes from. $\endgroup$
    – gringer
    Jul 8 at 13:57