MAGERI: Computational pipeline for molecular-barcoded targeted resequencing

Abstract

Unique molecular identifiers (UMIs) show outstanding performance in targeted high-throughput resequencing, being the most promising approach for the accurate identification of rare variants in complex DNA samples. This approach has application in multiple areas, including cancer diagnostics, thus demanding dedicated software and algorithms. Here we introduce MAGERI, a computational pipeline that efficiently handles all caveats of UMI-based analysis to obtain high-fidelity mutation profiles and call ultra-rare variants. Using an extensive set of benchmark datasets including gold-standard biological samples with known variant frequencies, cell-free DNA from tumor patient blood samples and publicly available UMI-encoded datasets we demonstrate that our method is both robust and efficient in calling rare variants. The versatility of our software is supported by accurate results obtained for both tumor DNA and viral RNA samples in datasets prepared using three different UMI-based protocols.

smCounter2: an accurate low-frequency variant caller for targeted sequencing data with unique molecular identifiers

“…The limitation of these algorithms is the requirement of many control samples for the site-specific error modeling. As an alternative, MAGERI [16] assumes a universal Beta distribution for all sites, which may result in lower accuracy compared to site-specific error modeling, but as a trade-off requires only one control sample, if the UMI coverage is high enough to observe the background errors and enough sites are covered to reveal the full distribution of error rates.…”

“…A two-step UMI-based variant calling approach that first constructs a consensus read with tools like fgbio [11] and then applies one of the conventional low-frequency variant callers [12] to the consensus reads has been implemented in [3, 13]. In addition to the two-stage method, three UMI-based variant callers, DeepSNVMiner [14], smCounter [15], and MAGERI [16], are publicly available. DeepSNVMiner relies on heuristic thresholds to draw consensus and call variants.…”

Family reunion via error correction: an efficient analysis of duplex sequencing data

“…However, reference-based approaches are inevitably biased and it was our main impetus to avoid the use of a reference sequence [13]. Alternatively, a strategy implemented in MAGERI, a tool which does not require a reference sequence to form consensus sequences, is able to perform efficient barcode error correction with the use of a custom seed-and-extend alignment algorithm [10,11]. However, it only forms singlestrand consensus sequences, not the duplex consensus sequences required in our analysis.…”

Family reunion via error correction: An efficient analysis of duplex sequencing data

“…Alternatively, a strategy implemented in MAGERI, a tool which does not require a reference sequence to form consensus sequences, is able to perform efficient barcode error correction with the use of a custom seed-and-extend alignment algorithm (Shugay et al 2017(Shugay et al , 2014 . However, it only forms single-strand consensus sequences, not the duplex consensus sequences required in our analysis.…”