The initial experimental design allowed us to generate very high sequence coverage per sample (approximately 24,000-fold). A more cost-effective approach would multiplex samples using DNA barcodes, which results in a lower effective coverage. In order to estimate the effect of a reduced coverage depth on the assay performance, we sampled the reads from the full coverage from one calibration (CAL-B) and two tested (CAL-C and CAL-D) samples to simulate multiplexing levels of 2, 4, 8, 16 and 32 samples per lane corresponding to a coverage depth around 12,000×, 6,000×, 3,000×, 1,500× and 750×, respectively (Table S6 in Additional file 2). The sensitivity remains above 92% at approximately 3,000× coverage or higher (Figure 1e) but drops to 85% at 1,500× coverage depth. In contrast, the PPV remains above 96.9% at all coverage depth. Furthermore, the estimation of the prevalence remains accurate (Figure S5 in Additional file 1). Most of the false negatives were variants expected at low prevalence for which a reduced coverage leads to few reads supporting the alternative allele. Thus, these data show that the performance of the UDT-Seq assay is maintained at an average coverage of approximately 3,000× and greater.
