A1041
Title: High-dimensional MANOVA test for semicontinuous biomedical data: Methodology and applications
Authors: Elena Sabbioni - University of Oxford (United Kingdom) [presenting]
Claudio Agostinelli - University of Trento (Italy)
Alessio Farcomeni - University of Rome Tor Vergata (Italy)
Elena Sabbioni - University of Oxford (United Kingdom)
Abstract: Modern biological applications increasingly involve complex data structures, driven by advances in technology that allow for the collection of a growing number of high-resolution features. In this context, the focus is on semicontinuous high-dimensional data, a common data type found in various fields, including genetics and medical research. These data combine a continuous part with positive observations and a part with exactly zero components, and often feature more variables than observations. This setting poses challenges for standard statistical methods, which typically address either semicontinuous or high-dimensional data, but not both simultaneously. To address this methodological gap, a novel MANOVA testing procedure is proposed, specifically designed to handle both of these features simultaneously. The method is based on a regularized likelihood ratio test, where the form of the penalized likelihood enables closed-form estimators, ensuring both computational tractability and scalability. Since the null distribution of the test statistic is unknown in this framework, we employ a permutation scheme. The efficiency of the method, both in terms of level and power of the test, is achieved in a simulation setting. Finally, its practical utility is illustrated through the analysis of a real microRNA expression dataset, showcasing its relevance for complex biomedical data.
