Simple statistical identification and removal of contaminant sequences in marker-gene and metagenomics data
Nicole M. Davis, Diana M. ProctoriD, Susan HolmesiD, David A. RelmaniD, Benjamin J. CallahaniD
Abstract
BackgroundThe accuracy of microbial community surveys based on marker-gene and metagenomic sequencing (MGS) suffers from the presence of contaminants -- DNA sequences not truly present in the sample. Contaminants come from various sources, including reagents. Appropriate laboratory practices can reduce contamination, but do not eliminate it. Here we introduce decontam (https://github.com/benjjneb/decontam), an open-source R package that implements a statistical classification procedure that identifies contaminants in MGS data based on two widely reproduced patterns: contaminants appear at higher frequencies in low-concentration samples, and are often found in negative controls.\n\nResultsdecontam classified amplicon sequence variants (ASVs) in a human oral dataset consistently with prior microscopic observations of the microbial taxa inhabiting that environment and previous reports of contaminant taxa. In metagenomics and marker-gene measurements of a dilution series, decontam substantially reduced technical variation arising from different sequencing protocols. The application of decontam to two recently published datasets corroborated and extended their conclusions that little evidence existed for an indigenous placenta microbiome, and that some low-frequency taxa seemingly associated with preterm birth were contaminants.\n\nConclusionsdecontam improves the quality of metagenomic and marker-gene sequencing by identifying and removing contaminant DNA sequences. decontam integrates easily with existing MGS workflows, and allows researchers to generate more accurate profiles of microbial communities at little to no additional cost.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 3,494
- FWCI
- 67.83
- Field percentile
- top 1%
- Organism
- Assay
- Research Focus
- Approach
In-depth summary
Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (25 Jul 2018) · CC BY 4.0
decontam, an R package, classifies contaminant sequences in marker-gene and metagenomic data by their inverse frequency-DNA concentration relationship and higher prevalence in negative controls.
External contaminant DNA, much of it from reagents, inflates diversity, blurs between-sample differences and creates spurious associations, especially in low-biomass samples. Common in silico fixes (abundance cutoffs, deleting anything seen in controls, blacklists) drop true taxa or miss abundant contaminants; decontam classifies each feature statistically without a prior contaminant list, while SourceTracker only estimates source proportions.
Key results
ShownIn a human oral 16S dataset, decontam scores largely agreed with curated genus-level reference lists of contaminant and oral genera. Per-ASV sensitivity was modest, per-read accuracy was much higher (near-perfect for the frequency and combined methods), and discriminative power rose with feature prevalence.
712 oral mucosa samples plus 33 extraction negative controls; at default P*=0.1, per-ASV sensitivity/specificity 0.33/1.00 (frequency), 0.61/0.95 (prevalence), 0.45/0.99 (combined); per-read 0.97/1.0000, 0.77/0.9994 and 0.98/0.9999 respectively. · decontam discriminates likely contaminants from…; Construction of oral and contamination databases
ShownOn the Salter et al. Salmonella bongori dilution series, the frequency method caught most contaminant reads in both 16S amplicon and shotgun data and never flagged the true organism. Scoring each batch separately worked better than pooling all batches.
Default P*=0.1: over 50% of contaminant amplicon reads and over 80% of contaminant shotgun reads classified as contaminants; no S. bongori reads flagged at any P*, S. bongori scores >0.98. Independent evaluation (Karstens et al., 8-strain mock dilution series): 74-91% of contaminant reads caught, no false positives. · Application of decontam to a dilution-series…
SuggestedRemoving contaminants that decontam flagged reduced batch effects between sequencing centers (16S) and DNA extraction kits (shotgun). The biggest gains came at intermediate dilutions, where both S. bongori and contaminants made up a substantial share of reads.
Bray-Curtis MDS ordination: between-batch sample distances decreased as P* rose from 0.0 (no removal) to 0.1 (default) to 0.5 (aggressive); three sequencing centers and three extraction kits, six samples per batch; assessed visually, with no quantitative distance statistic reported. · Application of decontam to a dilution-series…; Figure legends
ShownApplied through isNotContaminant to Lauder et al. placenta biopsy 16S data, the prevalence method found almost no candidate placental bacteria. Most putative non-contaminants were human DNA from off-target amplification, consistent with the data offering little evidence for an indigenous placental microbiome.
Of 810 ASVs present in at least five samples, six were called non-contaminants at FDR 0.5; five of those six matched Homo sapiens, and the remaining one was a Ruminococcaceae variant (a gut-associated taxon) whose true origin could not be established. · Identification of non-contaminant sequences in a…; Tables
ShownIn a gestational vaginal microbiota and preterm birth (PTB) dataset, batch-aware combined scoring flagged several of the ASVs most strongly associated with PTB as contaminants. All of them belong to genera previously reported as contaminants.
With sequencing runs as batches, four of the ASVs most significantly associated with PTB received combined scores below 10^−6; their genera (Herbaspirillum, Pseudomonas, Tumebacillus, Yersinia) appear in prior contaminant reports. The combined method gave the clearest low-score peak, and a 0.01 threshold suited this dataset. · Reduction of false-positive associations between…; Choice of classification threshold; Figure legends
Methods, data and reproducibility
Two de novo per-feature classifiers: a frequency model fitting log frequency vs log total DNA with slope −1 (contaminant) or 0 (non-contaminant), and a prevalence model comparing presence in negative controls vs true samples (chi-square or Fisher's exact). Scores can be computed per batch, frequency and prevalence scores merged via Fisher's method, and features below P* (default 0.1) called contaminants.
- Human oral mucosa 16S rRNA V4 amplicons (MiSeq) with reagent-only or blank-swab extraction negative controls; final table 18,285,750 reads, 2,420 ASVs, 767 samples · human (oral microbiome) · n = 712 oral mucosa samples; 33 negative controls · SRP126884
- Salter et al. Salmonella bongori monoculture dilution series, 16S rRNA amplicons from three sequencing centers; reported 16S qPCR used as DNA concentration proxy · Salmonella bongori (plus contaminants) · n = six samples per batch
- Salter et al. S. bongori dilution series, shotgun metagenomics with four DNA extraction kits (one excluded for low yield); no negative controls sequenced · Salmonella bongori (plus contaminants) · n = six samples per batch
- Lauder et al. placenta biopsy 16S rRNA data with multiple negative controls processed using two DNA extraction kits · human · n = 810 ASVs in at least five samples
- Callahan & DiGiulio et al. gestational vaginal microbiota ASV data on preterm birth, Stanford and UAB cohorts, sequencing runs treated as batches · human
- Curated oral and contamination genus databases (literature, microscopy, HOMD) used as reference ground truth for accuracy assessment · bacteria (genus-level lists)
- Methods & tools
- decontam R package (isContaminant, isNotContaminant), DADA2 1.8.0, QIIME split_libraries_fastq.py, phyloseq, SILVA nr v128 training set, Human Oral Microbiome Database (HOMD) 16S RefSeq V14.5, NCBI BLAST, PicoGreen Quant-iT dsDNA Assay, PowerSoil-HTP 96 well Soil DNA Isolation Kit, Illumina MiSeq v3, Bray-Curtis MDS ordination, Chi-square test, Fisher's exact test and Fisher's method
- Data & accessions
- https://benjjneb.github.io/DecontamManuscript/https://www.arb-silva.de/http://www.homd.org/
- Software & environment
- decontam distributed via GitHub and Bioconductor; DADA2 version 1.8.0; silva_nr_v128_train_set.fa; HOMD_16S_rRNA_RefSeq_V14.5.fasta; phyloseq; analyses provided as R markdown scripts.
- Reproducibility gaps
- Only the oral dataset has an accession (SRP126884) in the text; the Salter, placenta and preterm birth datasets are reused from cited studies without accessions or URLs given here.
Limitations (6)
- decontam assumes contaminants are distinct from true community members, so cross-contamination between processed samples (including abundant true sequences leaking into negative controls) is outside its scope. · stated by the authors
- Detection relies on cross-sample patterns, so contaminants present in very few samples are poorly caught (per-ASV sensitivity 0.33 to 0.61 at P*=0.1 in the oral data); the authors suggest pairing decontam with removal of low-prevalence features. · stated by the authors
- The frequency method breaks down when contaminants are a large share of reads (C ~ S or C > S), or when biomass differs systematically between sample groups or covaries with the condition of interest (e.g. mixed stool and airway samples). The authors advise running decontam separately per environment. · stated by the authors
- Prevalence-method sensitivity is limited by the number of sequenced negative controls. A simulation suggests 5-6 controls suffice, assuming true samples are far more numerous and prevalence patterns resemble the oral dataset. · stated by the authors
- Accuracy estimates use incomplete genus-level reference lists as ground truth. For example, Peptococcus, a known oral colonizer missing from the oral database, was counted as a ground-truth contaminant and so showed up as an apparent false negative. · stated by the authors
- There is no quantitative head-to-head test against abundance cutoffs, control subtraction, blacklists or SourceTracker on the same datasets. The claimed advantages over these approaches are argued rather than measured. · noted in this summary
Who should read:Read this if you run 16S or shotgun microbiome studies, especially low-biomass ones, and need a principled in silico way to flag reagent contaminants using DNA concentrations or negative controls.
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- Metacontam: A Negative Control-Free Decontamination Method for Metagenomic Analysis
Junwoo Jo, Hyeyoung Lee, … Sunjae Lee·bioRxiv·Apr 29, 2026
- Controlling for contaminants in low biomass 16S rRNA gene sequencing experiments
Lisa Karstens, Mark J. Asquith, … Shannon K. McWeeney·mSystemsPeer-reviewed·Jun 3, 2019
- Mitigation and detection of putative microbial contaminant reads from long-read metagenomic datasets
Stefany Ayala-Montaño, Ayorinde O. Afolayan, … Sandra Reuter·Microbial GenomicsPeer-reviewed·Jan 22, 2026
- Reagent contamination can critically impact sequence-based microbiome analyses
Susannah Salter, Michael J Cox, … Alan W Walker·bioRxiv·Jul 16, 2014
- Squeegee: de-novo identification of reagent and laboratory induced microbial contaminants in low biomass microbiomes
Yunxi Liu, Ryan A. Leo Elworth, … Todd J. Treangen·bioRxiv·May 6, 2021
- Targeted decontamination of sequencing data with CLEAN
Marie Lataretu, Sebastian Krautwurst, … Martin Hölzer·NAR Genomics and BioinformaticsPeer-reviewed·Jul 4, 2025