Contributions to biostatistics: categorical data analysis, data modeling and statistical inference
Mathieu Emily · HAL (Le Centre pour la Communication Scientifique Directe) · 2016
The recent years have seen biostatistics facing issues regarding the modeling and the analysis of more and more complex data. The evolution of biological data has been paved by two main technological revolutions: the explosion of computer capacities and the advent of high-throughput technologies. Although the fast evolution of biotechnologies has allowed the collection of massive amount of data, it has raised a large number of open questions. In the last few years, the deep modification of biological data type has contributed to the emergence of novel statistical challenges. In this context, the main goal of my research is to provide, in response to a biological question of interest, statistical procedures based on four main challenges: the design and the experimental planning to optimize statistical power, the statistical modeling of the types of measured variables, the formalization of relevant biological assumptions and the modeling of the structure of the data. My approach can be described through three main research axes: (1) the analysis and the modeling of categorical data, (2) the statistical modeling of highly structured data and (3) the probabilistic modeling and the statistical inference of spatial and temporal data. My contributions especially tackle the issue of association testing through the estimation of power functions, the detection of interaction, the variable selection and the correction for multiple testing. I have also focused my research activities on classification issues, the aggregation of statistical tests and the estimation of survival function. A systematic evaluation of the proposed methods has been performed through the analysis of real data from many fields of biosciences, such as genomics, proteomics, cancer, ecology and health. The results open novel perspectives in biostatistics, including the integration of heterogeneous data and personalized medicine.