Optimizing imputation of marker data from genotyping-by-sequencing (GBS) for genomic selection in non-model species: a case study of rubber (Hevea brasiliensis)

Genotyping-by-sequencing (GBS) is an efficient approach to get the high marker density required for genomic predictions (GP). However, GBS gives a relatively high proportion of sporadic missing SNP data and genotyping errors. For species without a high-quality reference genome with a chromosome level assembly, imputation of missing SNP data, mandatory for GP, requires methods that do not need the SNP physical positions. We compared, in terms of GP accuracy, seven map independent and two map dependent imputation approaches, and the use of all SNPs obtained by GBS against the subset of SNPs positioned on genetic maps.