Comparing high-dimensional data : interpretable two-sample testing by variable selection