MLPHENOM · Machine learning for quantitative modelling of structured phenotypes

Understanding phenotypic variation, and more particularly identifying the causal genetic or environmental regulators, is a major aim in biological investigations. The goal of this proposal is to develop and apply machine learning techniques to model key aspects of structure that occur in modern, high-dimensional phenotype datasets. First, the temporal structure of phenotypes that are recorded over time is addressed. Statistical models can exploit smoothness of time series and detect change points. Second, the structure of images, arising when digital pictures are used as phenotypic variables, is considered. Machine learning techniques allow interpretable image features to be automatically extracted and used as quantitative traits, complementing classical measurements. Finally, the network structure of the phenome is addressed. Different phenotype variables influence each other, resulting in a chain of effects that needs to be modelled to reveal the true causal relationships. The developed algorithms will be applied to understand phenotypic variation in Arabidopsis thaliana in direct collaboration with researchers at the Max Planck Institute for Developmental Biology.