As one of the cornerstones of predictive medicine, predictive genomics affords the capability to forecast a person’s risk for heritable disease and to anticipate inappropriate therapeutic options or dosing based on differences in how drugs are metabolized and absorbed among different individuals. To lucubrate the practical issue of identifying useful traits, genome wide association studies are typically undertaken. Identification of single nucleotide polymorphisms (SNP’s) are the bedrock of such studies whose aim is to identify causal relationships between genetic factors and complex diseases. Likewise, drug responses vary widely in populations with genetic factors estimated to contribute anywhere between 20 to 90% to the observed variability across cohorts. Thus, predictive genomics lies at the nexus of disease biology, personalized medicine, and translational bioinformatics. In order to provide a better understanding of how these technologies compare, we evaluate genome-wide imputation and variant calling performance using both low-pass whole genome sequencing (lpWGS) and genotyping microarrays.

Our data set includes 60 study samples of European and African ancestries from the 1000 Genomes project. For microarray data we assessed imputation performance using the pan-ethnic Applied Biosystem^TM Axiom^TM PMD array. For lpWGS we used publicly available data at 0.5x and 1x coverage from Gencove. For variant performance analysis, we used 44 samples of diverse ancestries with known pathogenic variants associated with severe Mendelian conditions or pharmacogenomic relevance.

We show the PMD array has imputation accuracy comparable to 1x lpWGS for the European cohort, and between 0.5x and 1x lpWGS for the African cohort. For the 38 known pathogenic variants, the array provided 100% non-reference allele concordance (NRC) across all MAF ranges in 144 carriers and 25 affected individuals. The NRC using lpWGS ranged between 95.1% (0.5x) and 97.6% (1x) for the common variants, between 87.5% (0.5x) and 93.8% (1x) for low frequency variants and between 73.3% (0.5x) and 96.7% (1x) for the rare variants. For the pharmacogenomic variants, PMD array showed an overall NRC of 99.6±4.9 and lpWGS showed NRC of 96.1±12.8 and 97.3±10.4 for 0.5x and 1x respectively.

Our results show that rare and low frequency variants as well as variants in complex genomic regions cannot be imputed with high accuracy using lpWGS at a depth of 1x in contrast to microarrays.

Jeanette Schmidt

Appearances