Genomic Selection Meets Transcriptomics: Predicting Quantitative Traits in Cassava

Quantitative traits are affected by many loci, whereas traditional allele-mining approaches operate on few loci at a time. To more effectively mine many alleles and increase their frequency in parallel, we use bioinformatic and statistical approaches borrowed from quantitative genetics. Genomic Selection (GS) methods have been developed to use all markers available across the genome. Previous studies in Cassava showed that even when using a relatively small training population and low-density GBS markers, GS can achieve reasonable prediction accuracies. We are currently interested in using functional annotation to tag SNPs that may be close to genomeregions that have some biological relevance for the trait of interest. As a case study we used data from an experiment evaluating the transcriptome of resistant and susceptible cassava plants after being infected with Cassava Brown Streak Virus (CBSV). Both a resistant and a susceptible cassava genotypes were evaluated at several time points after infection. WGCNA analysis allowed us to partition the set of differentially expressed genes based on their co-expression patterns across the different time points. We used the RNAseq data and in-silico genome wide identification of immunity related genes as input for a multikernel GS model approach. Prediction accuracies were compared between the traditional and the “informed” GS models using cross-validation.