Long-read amplicon denoising

Abstract

† These authors contributed equally Long-read next generation amplicon sequencing shows promise for studying complete genes or genomes from complex and diverse populations. Current long-read sequencing technologies have challenging error profiles, hindering data processing and incorporation into downstream analyses. Here we consider the problem of how to reconstruct, free of sequencing error, the true sequence variants and their associated frequencies. Called "amplicon denoising", this problem has been extensively studied for short-read sequencing technologies, but current solutions do not appear to generalize well to long reads with high indel error rates. We introduce two methods: one that runs nearly instantly and is very accurate for medium length reads (here ∼2.6kb) and high template coverage, and another, slower method that is more robust when reads are very long or coverage is lower. On one real dataset with ground truth, and on a number of simulated datasets, we compare our two approaches to each other and to existing algorithms. We outperform all tested methods in accuracy, with competitive run times even for our slower method.Fast Amplicon Denoising (FAD) and Robust Amplicon Denoising (RAD) are implemented purely in the Julia scientific computing language, and are hereby released along with a complete toolkit of functions that allow long-read amplicon sequence analysis pipelines to be constructed in pure Julia. Further, we make available a webserver to dramatically simplify the processing of long-read PacBio sequences.

Comparative Genomics and Full-Length TprK Profiling ofTreponema pallidumsubsp.pallidumReinfection

“…Our results indicate long-read sequencing is remarkably consistent with short-read sequencing for detecting TprK variants, consistent with the increasing accuracy of long-read sequencing [48]. However, as long-read sequencing is still less accurate than short-read sequencing [25,49], we could not confidently call TprK variants present at lower frequencies (< 0.2%). Increasing sequencing depth may help resolve TprK variants present at frequencies less than 0.2%.…”

“…As long-read sequencing often produces more errors than short-read sequencing, we used a read clustering-based denoising approach [25] to identify probable sequences of the tprK amplicons for each isolate. We filtered Q20 PacBio CCS reads based on size (1,400 - 1,800 bp) prior to applying denoising and then required each cluster to contain a minimum of 5 reads.…”

“…tprK sequencing reads were analyzed using custom Python/R scripts which are available on GitHub (https://github.com/greninger-lab/tprk). PacBio reads between 1400-1800 bp were trimmed of PCR primers using the dada2 preprocessing pipeline and denoised using RAD [25,26]. Illumina reads were trimmed using trimmomatic as described above.…”

High-throughput amplicon sequencing of the full-length 16S rRNA gene with single-nucleotide resolution

“…We predict that a computational workflow based on the DADA2 method will continue to be effective for PacBio CCS amplicons extending out to ~3 kb but that sensitivity to low-frequency variants will degrade for >3 kb amplicons as the fraction of error-free reads declines. In the regime in which few or no sequences are expected to be error-free, alternative computational methods will be necessary and are now being developed (Kumar 2019). We urge caution in applying our computational methods to data from PacBio RSII sequencing chemistries before P6-C4 and/or CCS data that was generated by the earlier SMRT Portal software, as error rates in such data may be substantially higher than in the data considered here.…”

A highly multiplexed droplet digital PCR assay to measure the intact HIV-1 proviral reservoir

“…We also classified all PSD sequences as intact and defective by sequencing, using criteria described by the Pro-Seq IT tool associated with the PSD database, which include intactness thresholds for sequence length, mutations, and deletions. 13,36,37 We then compared the two classifications (ddPCR and sequencing) to quantify agreement. Of the 1,071 PSD sequences, 966 sequences (90.2%) agreed between the PSD algorithm and our ddPCR protocol (Figure 4A).…”