Figure 3

SNP metrics for different multiplexes and analysis strategies (third stage of our multiplexed t NGS test model). This figure summarizes the SNP detection and genotype concordance results of the human BRCA1 and BRCA2 experiments. The results are detailed for known SNPs within the TR for the 26 barcoded Yoruban samples (two 4-plexes, one 8-, 16- and 20-plex) and the non-barcoded control. The results shown in this figure were generated using Bioscope 1.0.1 (mapping, SNP-calling), SAET (pre-mapping read enhancement), SAMtools (SNP-calling), pibase (SNP-typing), and IGV (manual inspection). The identified SNPs were benchmarked against a ‘silver’ and a ‘gold’ consensus of published and validated genotypes (see Methods section). Known SNPs which we failed to detect with a specific tool are counted as false negatives (FN, dark grey for the ‘silver’ consensus and brown for the ‘gold’ consensus). Potential novel SNPs which could not be validated with certainty are counted as false positives (FP, blue for SNPs in only one library and light blue for SNPs in several libraries). The first four groups of columns show the automatic SNP-calling results, and the rightmost two groups show the manual inspection results. For TR mapping (first two groups), Bioscope detected more known SNPs than for whole-genome mapping (middle groups, no barcode, 4-plex) and also more FP SNPs. With SAET read enhancement Bioscope detected fewer known SNPs (second and fourth group, TR 4-plex and no barcode, WG 4-plex) than without SAET (first and third group). The SAMtools SNP-caller (fifth and sixth groups) performed worse than Bioscope. Our pibase re-analysis (seventh group) and manual inspection in IGV (rightmost two groups) revealed that the Bioscope and SAMtools SNP-callers filtered out known SNPs which were detected with pibase and also seen in the mapped reads. A detailed description is given in the Results section. In summary, to automatically detect SNPs with minimal manual filtering of false positives or false negatives, and to take advantage of short run times, we recommend a combination of multiplexing technical replicates, mapping to the TR, and backmapping the reads covering the SNPs to the WG (see Conclusion section).