Data-adaptive statistics for multiple hypothesis testing in high-dimensional settings