DADA2: High resolution sample inference from amplicon data

Abstract

Microbial communities are commonly characterized by amplifying and sequencing target genes, but errors limit the precision of amplicon sequencing. We present DADA2, a software package that models and corrects amplicon errors. DADA2 identified more real variants than other methods in Illumina-sequenced mock communities, some differing by a single nucleotide, while outputting fewer spurious sequences. DADA2 analysis of vaginal samples revealed a diversity of Lactobacillus crispatus strains undetected by OTU methods.The importance of microbial communities to human and environmental health has motivated methods for their efficient characterization. The most common, and cost-effective, method is the amplification and sequencing of targeted genetic elements. Amplicon sequencing of taxonomic marker genes such as 16S rRNA [1], the ITS region [2] or 18S rRNA [3] provides a census of a community. Functional diversity can be probed by targeting functional genes [4].Disentangling errors from biological variation in amplicon sequencing data presents unique challenges, which has prompted the development of amplicon-specific error-correction methods [5,6,7,8]. Most of these methods were designed for pyrosequenced amplicons, and cannot be applied to Illumina sequencing.Currently, errors in Illumina-sequenced amplicon data are most often addressed by filtering low quality reads and constructing Operational Taxonomic Units (OTUs): clusters of sequences that differ by less than a fixed dissimilarity threshold (typically 3%) within which sequence variation is ignored [9,10,11]. Lumping similar sequences together reduces the rate at which errors are misinterpreted as biological variation, but OTUs under-utilize the quality of modern sequencing by precluding the possibility of resolving fine-scale (or strain-level ) variation [7,12,13,14,15]. Recent studies have shown that fine-scale variation can be informative about ecological niches [12,13], temporal dynamics [15], and population structure [4]. Fine-scale variation differentiates pathogenic from commensal strains in some cases [16,17], and can contain clinically relevant information for more complex microbiome-associated diseases [18,19,20].DADA -the Divisive Amplicon Denoising Algorithm -was introduced to correct pyrosequenced amplicon errors without constructing OTUs [7]. DADA was shown to identify real variation at the finest scales in 454-sequenced amplicon data while outputting few false positives [7,4].Here we present DADA2, an extension and reimplementation of DADA adapted for use with Illumina sequencing and available as an open-source R package available at https: //github.com/benjjneb/dada2. DADA2 implements a new model of Illumina-sequenced amplicon errors that incorporates quality information. Banded alignments and a kmerdistance screen improve computational performance. The DADA2 R package provides light-weight tools for other key parts of the amplicon denoising workflow: filtering, derepli-1 cation, chimera identification, and merging paired-end reads.We compared DAD...

Piphillin: Improved Prediction of Metagenomic Content by Direct Inference from Human Microbiomes

“…It can also be used in conjunction with Mothur [31], RDP [32], DADA2 [33] or UPARSE [34]. Piphillin requires only two input files from the user, an OTU abundance table and a fasta file with OTU representative sequences.…”

Improvement of identification methods for honeybee specific Lactic Acid Bacteria; future approaches

“…Since the DADA2 program retrieves biological sequences from reads by modeling the Illumina-sequencing errors [32], and different sequencing runs might have distinct qualities, we ran the DADA2 pipeline (v.1.1.1, http://benjjneb.github.io/dada2/tutorial.html) on our samples from the two runs separately. After examining the quality profiles of forward and reverse reads, we trimmed off the low-quality tails from the 3’ ends to make the filtered reads having an average Phred score Q >30 (the lengths of truncated forward and reverse reads were 230 and 150 bp, respectively), meanwhile, we required no N (maxN = 0), no PhiX sequences (rm.phix = TRUE) and the maximum number of expected error being 2 (maxEE = 2) for all paired-end reads.…”

“…Until recently, a significant challenge of Illumina rRNA gene amplicon sequencing has been to accurately identify true biological sequence variants while excluding amplicon sequencing errors [32–34]. A common way to study microbial communities is to cluster similar amplicon reads, which masks most amplification errors [33,35,36], but at the same time removes some true biological variants.…”

“…Consequently, the aim of the present study was to demonstrate the requirements for vital growth supplements of the 13 hbs-LAB strains by monitoring growth rates. Another objective was to evaluate a novel identification approach to differentiate the hbs-LAB strains using a recently published Illumina amplicon error-correction program [32] for retrieving 16S rRNA gene sequences from Illumina sequences.…”

Reproducible Research Workflow in R for the Analysis of Personalized Human Microbiome Data

“…We focus here on two Bioconductor 5 packages—dada2 6 and phyloseq 7,8 —created specifically to analyze amplicon sequencing data within the R environment, and show how they enable reproducible research in several microbiome studies. We begin by illustrating the need for reproducible research workows in microbiome studies with a typical workow example.…”