BCAM Scientific Seminar Statistical modelling of genotype to phenotype relations

A central problem in plant genetics is to predict the phenotype, an observable trait, from genetic and environmental information. The data for this problem typically come from series of designed field experiments in which a panel of genotypes (plants with a defined genetic constitution) is evaluated for a number of phenotypic traits. Some examples of phenotypic traits are yield, plant height, earliness, and disease resistance. For many research objectives in plant genetics, it is useful to predict traits from DNA variation, i.e. single nucleotide polymorphisms (SNPs). Prediction has become challenging as the phenotypic responses include 100 to 2000 genotypes with dependencies between them, with observations coming from multiple experiments under different environmental conditions, while the predictor set covers between 1000 to 10 million SNPs. Linear mixed models and penalized regressions will be presented with attention for the modelling of relationships between genotypes and between environmental conditions, and the interactions between genetic factors (SNPs in genes) and environmental factors (water availability, temperature). Some extensions to multi-trait models and graphical models will be demonstrated. 

Joint work with Daniela Bustos-Korts, Emilie Millet, Martin Boer, Marcos Malosetti and Willem Kruijer.