We show that, as compared to other commonly used methods, our algorithm can achieve a higher true positive rate whilst controlling a false discovery rate of 1%.