New insights into the usefulness of robust singular value decomposition in statistical genetics

The distribution of continuous real life variables is usually not normal and plant phenotypes are no exception to the rule. These distributions often show heavy tails which are sometimes asymmetric. In such scenarios, the classical approach whose likelihood-based inference leans on the normality assumption may be inappropriate, having low statistical efficiency.Moreover, association tests may also be underpowered. Robust statistical methods aredesigned to accommodate for certain data deficiencies, allowing for reliable results under various conditions. They are designed to be resistant to influent factors as outlying observations,non-normality and other model misspecifications. Additionally, if the model verifies the classical assumptions, robust methods provide results close to the classical ones. Therefore, a new methodology where robust statistical methods replace the classic ones to model, structure andanalyse genotype-by-environment interactions in the context of multi-location plant breeding trials,is presented. Here interest lies in the development of a robust version of the additive main effects and multiplicative interaction model whose performance is compared with its classical version. This is achieved through Monte Carlo simulations where one particular contamination scheme is considered.