Learning predictive vine copula models for complex plant traits



Securing stable food supplies for a growing human population is required for survival. Genetic improvement of crops is key for achieving this goal. Efficient plant breeding aims at predicting plant performance from genomic data to generate crops adapted to future requirements. However, the genomic data is high-dimensional, and plant traits are measured in different environments for various populations. Thus, learning computationally tractable and accurate predictive models that can capture flexible dependence structures, interactions in multi-trait, multi-environment, and multi-population data is required. We suggest to approach these tasks with the class of vine copula models to develop more accurate prediction models for multiple traits from genomic information. Such a vine copula based approach to genomic prediction does not exist so far.