Evaluating Metabochip Genotyping Accuracy

We evaluated the utility of the Metabochip and accuracy of its genotype calls in three sample sets: (1) 15,896 northern European individuals from the FUSION, METSIM, HUNT, Tromsø, and Diagen studies [26] (link)–[30] together with 67 HapMap samples genotyped at least two times each and called using Illumina GenomeStudio software by re-clustering these data; (2) 6,614 Sardinian individuals organized in 1,243 extended families from the SardiNIA study [31] (link), [32] called by GenomeStudio software using default cluster data; and (3) 9,715 Nordic individuals from the Malmø Preventive Project, the Scania Diabetes Registry, and the Botnia Study [33] (link)–[35] (link) genotyped using a modified version of the BIRDSEED genotype calling algorithm [36] (link).
We applied standard SNP- and sample-based QC filters based on call rate, Hardy-Weinberg equilibrium deviations, duplicate genotype inconsistencies, and failures of Mendelian inheritance; in the Nordic sample, we also carried out checks based on plate-specific characteristics. These filters resulted in final data sets of 163,222 polymorphic SNPs genotyped in 67 HapMap samples, 142,812 polymorphic SNPs genotyped in 6,164 Sardinians, and 179,165 polymorphic SNPs genotyped in 8,473 Nordic individuals.