Variant Prioritization for Genomic Selection in Cassava

With the advent of next generation sequencing Genomic prediction and Genome wide association studies are the leading approaches in the use of whole genome marker and phenotypic data. Results coming from these approaches have been considered independently from each other and used aiming at different goals.Through whole-genomic prediction genotypic or breeding values for both observed and unobserved individuals can be predicted. While GWAS studies aim to dissect the genetic architecture detecting markers significantly associated to traits with a defined threshold.

To date, prediction models in plants have used all markers across the genome for prediction assuming equal contribution and tend to avoid marker selection. In this scenario in Cassava moderate to low prediction accuracies values have been observed

In previous studies a generalized way to building trait-specific genomic relationships matrices, using GWAS results via BLUP, has been tested using animal and plant models.Additionally, statistical models can be improved if structural genome annotation is included with subsets of annotated SNPs as GRM-kernels.

Within the scope of the present study we included genome-wide association results (p-values and marker effects) of root weight, root number and as dry matter as weights for the single nucleotide polymorphism (SNP) markers when constructing the genomic relationship matrix (GRM) for genomic prediction. Moreover we followed a multi-kernel approach partitioning SNPs based on their statistical significance/or not in association mapping studies and biological/functional annotation.

Results and implications of the study in terms of genomic prediction accuracies are discussed .