Obtaining genome-wide genotype information for millions of SNPs in soybean [Glycine max (L.) Merr.] often involves completely resequencing a line at 5X or greater coverage. Currently, hundreds of soybean lines have been resequenced at high depth levels with their data deposited in the NCBI Short Read Archive. This publicly available dataset may be leveraged as an imputation reference panel in combination with skim (low coverage) sequencing of new soybean genotypes to economically obtain

more »

... sity SNP information. Ninety-nine soybean lines resequenced at an average of 17.1X were used to generate a reference panel, with over 10 million SNPs called using GATK's Haplotype Caller tool. Whole genome resequencing at approximately 1X depth was performed on 114 previously ungenotyped experimental soybean lines. Coverages down to 0.1X were analyzed by randomly subsetting raw reads from the original 1X sequence data. SNPs discovered in the reference panel were genotyped in the experimental lines after aligning to the soybean reference genome, and missing markers imputed using Beagle 4.1. Sequencing depth of the experimental lines could be reduced to 0.3X while still retaining an accuracy of 97.8%. Accuracy was inversely related to minor allele frequency, and highly correlated with marker linkage disequilibrium. The high accuracy of skim sequencing combined with imputation provides a low cost method for obtaining dense genotypic information that can be used for various genomics applications in soybean. KEYWORDS imputation high density SNP data skim sequencing low cost genotyping soybean Genomics research has yielded a variety of tools which allow for more efficient and precise translation of genetic variation into crop improvements. Panels of single nucleotide polymorphisms (SNPs) obtained through SNP arrays or genotyping-by-sequencing (GBS) are the most common tool used to explore and make associations between genetic and phenotypic variation. Genomics-assisted crop breeding continues to demand increasing densities of genotype information to successfully dissect and predict genetically complex traits (Hamblin et al. 2011; Lorenz et al. 2011) . Current approaches of directly ascertaining a high density of SNP genotype data on large populations are cost prohibitive or fall short of being able capture the maximum amount of genetic space. Fixed SNP arrays and GBS are popular options for SNP genotyping in crops. Panels ranging in densities of up to 600,000 variants are now common in several crop species (Rasheed et al. 2017 ). However, recent genomics studies are utilizing datasets consisting of one million or more markers to answer complex, quantitative genetic questions. The need for this high density of markers is rendering current arrays and GBS approaches inadequate to generate the magnitude of data modern genomic studies require (Tian et al. 2011; Patil et al. 2016; Li et al. 2018) . High-depth whole genome sequencing can achieve these marker densities. One study utilizing high-depth whole genome sequencing in soybean found 9,107,000 high quality SNPs (Valliyodan et al. 2016) . Despite advances and the plummeting cost of next generation sequencing (NGS) data, this approach still presents a heavy financial burden, as several reads are required at each variant site to ensure data quality and completeness. Decreasing genome coverage in the interest of cost savings introduces missing data, which decreases power and can produce biased results. Imputation of missing data has the potential to allow the researcher to recover nearly all of the missing data points resulting from skim sequencing, drastically reducing genotyping expenses associated generating complete, high quality, high resolution SNP datasets. By predicting the unobserved genotypes based on the surrounding