Improving nut and kernel quality traits is a high priority in almond breeding programs around the world. Almond has a long juvenile period and phenotypic selection for nut and kernel traits can only be conducted after three years from planting. In early stages of planting, individuals with desirable nut and kernel traits can be identified by marker-trait associations (MTAs) using molecular markers. Currently, MTAs are identified by quantitative trait locus (QTL) mapping using progeny from bi-parental crosses or association mapping panels. However, the efforts of identifying MTAs using current QTL detection methods are hampered either by unavailability of genomic information or required genetic linkage maps. In addition, most kernel traits have polygenic inheritance, and many genes and genomic regions affect genetic variations. In crop research, genomic selection would provide promising approach to accelerate the genetic gains and reduce the length of breeding cycle. Yet, application of genomic selection in almond breeding and research is limited. We present results demonstrating the predictive ability of whole-genome and pedigree-based models to identify elite candidate parents for almond kernel weight. In this work, we used ancestral pedigree and phenotypic data from 13,000 progeny that were derived from 57 parents and 291 families. Ancestral pedigrees were recorded from the available literature from the almond breeding programs in USA, Spain, Italy, France, and Australia. Average kernel weight was obtained for each progeny tree from 30 nuts. All parents were resequenced using whole-genome sequencing at a depth of 15x. Over 80k high quality, independent single-nucleotide polymorphisms were used to construct realised genomic relationship matrix and linkage disequilibrium (LD) regions were used to compute LD weights. Genomic best linear unbiased prediction (GBULP) using Asreml-R was used to predict genomic estimated breeding values (GEBV). Pedigree model derived from linear mixed model was used to predict individual tree effects (PEBV) to validate the predicted GEBVs. EBVs were compared using Pearson correlation coefficient (r) and elite candidate parents were selected based on the selection index. For kernel weight, both pedigree and genomic models resulted similar EBVs, and r was 0.97. A high level of correlation in EBVs obtained from two methods indicates the suitability of these models in estimating BVs for future predictions. Predicted elite candidate parents from this study can reduce the conventional breeding cycle of almond by 6 years. The constructed models mainly represent Australian context and multi-environmental trials are required to identify the broader applicability of these models.
