Scanning and Filling: Ultra-Dense SNP Genotyping in Soybean Combining GBS and Resequencing Data

Genotyping by sequencing (GBS) constitutes a highly cost effective approach for simultaneous SNP discovery and genotyping. By design, GBS performs a partial scan of the genome and this inevitably leads to missing data as the portion of the genome that is captured in each individual is not identical. We used GBS data on 301 soybean lines to explore how the amount of missing data allowed (MaxMD) affects the number of SNPs called and how accurately these missing data can be imputed. When increasing MaxMD from 20 to 80% incrementally, the number of SNPs called increased from 12,712 to 62,643. The mean amount of missing data under these conditions ranged between 7% (at MaxMD=20%) and 50% (MaxMD=80%). Missing data were imputed and compared to the genotype determined through whole-genome resequencing of 23 of the lines (9x depth, 96% coverage). Collectively, 3.6 million SNPs were identified among this subset of lines and almost all (98.4%) of the genotypes called by GBS proved accurate. Imputation of missing data was more accurate when performed with a larger number of SNP markers (albeit with more missing data). At MaxMD=80%, 94% of SNP genotypes were correctly imputed, whereas at MaxMD=20%, the accuracy decreased to 86%. Jointly, called and imputed markers were highly accurate (96-98%). Finally, 1.4 million tag SNPs from the sequenced lines were imputed across the entire population with greater than 90% accuracy. This ultra-dense marker coverage resulted in much stronger associations in genome-wide association work.