Fig. 2: Characterization of QBDA AML panel for minimal residual disease (MRD) detection.

a Limit of detection (LoD) threshold for different types of mutations. b Observed mutation VAF in a spike-in positive sample and a healthy PBMC sample. The positive sample was prepared by mixing Horizon Myeloid DNA Reference Standard, 3 synthetic gBlocks, and a gDNA sample extracted from healthy PBMC, resulting in VAF between 0.001% and 0.1% for 22 different mutations. Sixteen out of 22 mutations were around 0.01% VAF (between 0.005% and 0.02%). The “expected” VAF was quantitated by UMI-based NGS without mutation enrichment. All 22 mutations covered by the AML panel were observed in the positive sample (orange line); 82% of the mutations were within twofold of expected VAF. The same healthy PBMC sample was also analyzed alone as the paired negative sample using the AML panel (gray line). In a healthy sample, some mutations (C > T or G > A) are observed at below-LoD level, possibly due to clonal hematopoiesis. Here 1 µg of gDNA was used for each library. c Quantitation accuracy. The positive sample in (b) was sequenced in triplicate NGS libraries; two additional positive samples with threefold or fivefold VAF of the abovementioned sample were also analyzed. For each of the 22 mutations, the observed VAF was in correct order for the 1×, 3×, and 5× VAF samples. In the triplicate experiment of the 1× VAF (≈0.01%) sample, one mutation was not observed in one of the replicates, thus the sensitivity is approximately 1–1/(22 × 3) = 98.5%. One micrograms of gDNA was used for each library. d Sequencing depth down to 45,000× does not affect sensitivity in 1X VAF (≈0.01%) sample. The 1× VAF positive sample (500 ng input) was sequenced with 350,000× depth (7.7 M reads). Even after downsampling to 45,000× depth sequencing by random sampling 1.0 M reads from the original library, all mutations are observed. The median observed UMI counts from 20 independent simulations were plotted against observed UMI counts in the original library.