In the first part of my talk, I will give a brief overview of prediction modelling in the field of omics research. Particular attention will be given to important issues such as the appropriate handling of clinical and multi-omics data, the transportability of prediction models for application on independent data, the prevention of “fishing expeditions” and related questionable research practices, and the appropriate representation of uncertainties related to the data analysis strategy. In the second part of my talk, I will discuss problems related to the evaluation of prediction methods with examples from omics research. I will argue that good scientific practice principles proposed in the last years to address the general replicability crisis are not only relevant to data analysts working in biomedical projects, but also, in a perhaps more subtle and not very well-understood way, to methodological researchers (i.e. researchers working on the development of new data analysis methods) in their own research projects. Issues such as the publication bias, the design of benchmark experiments or fishing for significance in methodological statistical research will be illustrated through examples from prediction modelling, including a large-scale comparison study of prediction methods for multi-omics data.

Einladung von Prof. Vanessa Didelez