Effects of filtration on imputation in clusterised variants
Résumé
The standards based on the same principles of fi ltration are commonly used
for genome-wide association studies and imputations. To investigate the eff ect
of imputation with and without pre-fi ltration of SNP followed by post-fi ltration
on the variants imputed at a conservative and less stringent threshold, we
categorised them in diff erent classes. We used 1,031 individuals from diverse
ethnicities and compared also their allele frequencies with the 1,089 NCBI
recorded individuals, within a 2Mb of chr20 dbSNPB37.p13, after curation of
the variants database. Only the markers removed during quality control (QC)
and absent from the reference genome were not imputed, with shapeit 2 and
impute 2 . Hence, to maintain genotyped variants the pre-fi ltration could be less
stringent, but no signifi cant diff erences in that number was observed between
the imputation prior and after pre-fi ltration of SNP. High correlation between
frequencies minor alleles generated after imputation were found between both
conditions. We didn’t fi nd any signifi cative diff erences between the frequencies, except within the range of very rare and rare variants . However, the magnitude of those diff erences were small and became non signifi cative under
the hypothesis testing their true mean diff erences m = 1.85E-05. There was a slight loss of information after QC. The presence of low quality genotyped variants prior to imputation did not impair neither their imputation quality that
showed maximum, nor the imputation of the remaining good quality SNP.
When an indel and polymorphism were present at the same locus, there was
dual imputation. Variants with no repository names, with position for sole
identifi er produced unreliable maf. Null allele were detected in NCBI database,
by comparison with the imputed variants, and vice et versa, NCBI showed
records while many null alleles would be imputed. We also considered as
unreliable, SNP with null allele frequencies in both the imputed results and
the NCBI dbSNP B37.p13, having very poor quality scores. An addition of
0.5 in the post-fi ltration imputation score stringency from 0.3 to 0.8 lead to a
decrease in the number of SNV < 0.01 maf by 1.8 fold, in both conditions with
and without prior QC fi ltration, and by a 2.5 fold in the number of SNV < 0.001
maf and lowered by half the number of very rare variants (< 5E-04). With an
average maf > 0.01, a standard threshold, showed a mean score > 0.8 whether QC was performed or not.
Origine | Fichiers produits par l'(les) auteur(s) |
---|