We developed a four-step approach to identify low-prevalence mutations (Figure S4 in Additional file 1). In step 1 the sequencing error rate is estimated using invariant bases in a calibration sample; in step 2, the candidate mutations are filtered and their level of significance is determined in both calibration and tested samples using the error rate; in step 3, the significance threshold is calculated using the known SNPs from the calibration sample; finally, in step 4, the significant mutations are called in the tested sample using this threshold. Following this procedure we used all the CAL samples in turn for the calibration and testing, thus providing a comprehensive evaluation of the assay performance across multiple calibration-tested sample combinations and sequencing runs.
