Moreover, our study could also be extended to yield differentiated results for specific prediction tasks, e.g., prediction of disease outcome based on different types of omics data, or prediction of protein structure and function. In the present study, we intentionally considered a broad spectrum of data types to achieve a high number of datasets. Obviously, performance may depend on the particular prediction task, which should be addressed in more focused benchmark studies conducted by experts of the corresponding prediction task with good knowledge of the considered substantive context. However, the more specific the considered prediction task and data type, the more difficult it will be to collect the needed number of datasets to achieve the desired power. In real data studies, there is a trade-off between the homogeneity and the number of available datasets.
