Using the error rate generated from the calibration sample, we calculated the binomial test P-values and ranked the candidate position in the tested sample accordingly. We kept all positions above the P-value determined by the MCC threshold obtained in the calibration sample.