In search of rare variants: preliminary results from whole genome sequencing of 1,325 individuals with psychophysiological endophenotypes.

Schematic analysis overview. For additional details see Iacono et al., 2014.

Discordance rates between the integrated array genotypes and sequence genotypes. This plot provides a description of the accuracy of the genotype calls from whole genome sequencing. The bar chart along the bottom gives the fraction of genotypes that were homozygous reference (HomRef), heterozygous (Het), and homozygous alternate (HomAlt), for the full range of possible nonreference allele counts. For example, if an individual in the study was called homozygous reference on the array (i.e., homozygous for the same allele that exists on the reference genome GRCh37), then the red dots give the rate at which that individual was called something other than homozygous reference in the sequence data. For SNPs with a nonreference allele count of 1–10, the sequence error rate was approximately 1 in 10,000. For SNPs with nonreference allele count of 1,500 (MAF ~50%), the sequence error rate was approximately 1 in 1,000. Similarly, if an individual was called heterozygous for some SNP on the array, then the green dots give the rate at which individuals were called heterozygous in the sequences. For a site with nonreference allele count of 1–10, the sequence error rate was about 20%. (Note that this 20% is based on only 141 genotypes—individuals homozygous for an alternate allele are rare themselves.) For sites with nonreference allele counts of 1,500, the rate was a little over 1 in 1,000. In general, sequencing was highly accurate, with accuracy falling off for the rarest variants.

Imputation accuracy comparison between MTFS sequence and 1000 Genomes reference panels. This plot provides a comparison of the imputation accuracy on Chromosome 20 obtained with 1000 Genomes compared to MCTFR sequence. Plotted in solid lines is the squared Pearson correlation between the imputed dosage and the genotyped minor allele counts for a range of allele frequencies, using MCTFR sequences (black) and 1000 Genomes sequences (blue) as imputation reference panels. To make a direct comparison, the SNPs are restricted imputation to sites on the genome-wide 660W-Quad SNP array, and then tested accuracy on rare variants on the exome chip. The number of SNPs contributing to each window are given in red, and each window was centered on each dot in the red line. The plots show that the MTFS reference panel provides imputation results for all MAFs, and especially better results for less common alleles.