Bio Workflows Edna Pipeline

Version Compatibility

Reference examples tested with: DADA2 1.30+, FastQC 0.12+, MultiQC 1.21+, cutadapt 4.4+, phyloseq 1.46+, vegan 2.6+

Before using code patterns, verify installed versions match. If versions differ:

• R: packageVersion('<pkg>') then ?function_name to verify parameters
• CLI: <tool> --version then <tool> --help to confirm flags

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

eDNA Metabarcoding Pipeline

"Process my eDNA samples from raw reads to community ecology" -> Orchestrate primer removal, denoising (OBITools3 or DADA2), contamination filtering, taxonomy assignment, Hill number diversity estimation, and constrained ordination for species-environment analysis.

This is a workflow skill: it owns the chaining decisions and hand-offs, not the internals of any one step.

The governing principle

An eDNA community table is a position in a choice-chain (marker -> primer bias -> denoise -> decontam -> tag-jump -> taxonomy DB), never a census; the trustworthy result is decided at these seams.

1. The marker + primer set + reference DB are committed together (COI/12S/ITS/rbcL/18S). The marker fixes both what taxa amplify and the achievable assignment rate; 50-85% unassigned AT SPECIES LEVEL is TYPICAL (higher ranks assign far better) and must be reported honestly (a high unassigned fraction is a database gap, not a pipeline failure). Report the marker's known primer bias with every result.
2. Read counts are NOT abundance (Lamb 2019). eDNA read counts reflect biomass AND primer affinity AND copy number AND degradation — commit to presence/occupancy or effective-species-count framing, never raw-read "abundance". This is the eDNA analogue of the metagenomics read-fraction != cell-fraction seam.
3. Coverage-based (not size-based) rarefaction; Hill numbers as effective species counts; Chao1 is a LOWER bound. Commit to iNEXT coverage-standardization at C~0.95 and the doubling-rule extrapolation bound (Chao & Jost 2012); report Chao1 as ">=" and NEVER compute it after aggressive denoising (singletons stripped -> Chao1 degenerates to observed richness).
4. The platform sets the tag-jump baseline, committed at sequencing. The tag-jump phenomenon (Schnell 2015) has a platform-dependent rate: MiSeq index-hopping ~0.001-0.005 vs NovaSeq patterned flow cells ~10x higher; quantify and platform-filter the residual rate. And decontam is SCREENING, not a classifier (Davis 2018) — every flagged ASV gets a biological-plausibility review; retain only where statistics AND plausibility agree.

Made-once commitments

Pipeline Overview

Raw amplicon FASTQ (demultiplexed)
    |
    v
[1. QC] ------------------> FastQC / MultiQC quality assessment
    |
    v
[2. Primer Removal] ------> Cutadapt (remove forward + reverse primers)
    |                            |
    |                            +---> QC: reads per sample >1000
    |
    +--- Path A: OBITools3     +--- Path B: DADA2
    |       |                  |       |
    |       v                  |       v
    |   [3a. obi alignpairedend]   [3b. filterAndTrim]
    |       |                  |       |
    |       v                  |       v
    |   [4a. obi uniq]        |   [4b. learnErrors + dada]
    |       |                  |       |
    |       v                  |       v
    |   [5a. obi ecotag]      |   [5b. assignTaxonomy]
    |                          |
    +-------- Merge -----------+
                |
                v
[6. Contamination Filter] -> decontam / microDecon (negative control removal)
    |
    v
[7. Taxonomy Table] -------> Species x sample matrix
    |
    v
[8. Diversity Analysis] ---> iNEXT Hill numbers (q=0,1,2)
    |
    v
[9. Community Comparison] -> vegan CCA/RDA + indicspecies
    |
    v
Species table + diversity metrics + ordination plots

Step 1: Quality Assessment

fastqc -t 8 -o fastqc_output/ raw_reads/*.fastq.gz
multiqc fastqc_output/ -o multiqc_report/

Step 2: Primer Removal

# Adapter sequences are marker-specific; examples below for common eDNA markers
# --discard-untrimmed: remove reads without primers (likely off-target)
# --minimum-length 50: discard very short fragments after trimming

# COI (Leray primers mlCOIintF / jgHCO2198)
cutadapt -g GGWACWGGWTGAACWGTWTAYCCYCC -G TAIACYTCIGGRTGICCRAARAAYCA \
    --discard-untrimmed --minimum-length 50 -j 8 \
    -o trimmed/{sample}_R1.fastq.gz -p trimmed/{sample}_R2.fastq.gz \
    raw_reads/{sample}_R1.fastq.gz raw_reads/{sample}_R2.fastq.gz

Common primer sets by marker:

QC Checkpoint: Demultiplexing

# Gate: reads per sample >1000; negative controls <100 reads
for f in trimmed/*_R1.fastq.gz; do
    sample=$(basename "$f" _R1.fastq.gz)
    count=$(zcat "$f" | awk 'END{print NR/4}')
    echo "$sample: $count reads"
done

Step 3: Paired-End Merging and Denoising

Path A: OBITools3

# Import, align, and SAMPLE-TAG each demultiplexed sample separately, then concatenate.
# On multiplexed data `obi ngsfilter -t tagfile` assigns the `sample` tag; with pre-demultiplexed
# FASTQ it must be set explicitly (-S takes TAG:PYTHON_EXPRESSION), otherwise `obi uniq -m sample`
# has nothing to merge on and the MERGED_sample tag `obi clean -s` needs is never created.
for s in $(cat samples.txt); do
    obi import --fastq-input trimmed/${s}_R1.fastq.gz reads/${s}_r1
    obi import --fastq-input trimmed/${s}_R2.fastq.gz reads/${s}_r2
    obi alignpairedend -R reads/${s}_r2 reads/${s}_r1 reads/${s}_aligned
    obi annotate -S "sample:'${s}'" reads/${s}_aligned reads/${s}_tagged
done
obi cat $(for s in $(cat samples.txt); do printf -- '-c reads/%s_tagged ' "$s"; done) reads/aligned

# Filter by alignment score
# score >= 50: removes poorly overlapping pairs
obi grep -p 'sequence["score"] >= 50' reads/aligned reads/filtered

# Filter by merged length (marker-dependent range)
# 100-500 bp: typical COI amplicon range
obi grep -p 'len(sequence) >= 100 and len(sequence) <= 500' \
    reads/filtered reads/length_filtered

# Dereplicate
obi uniq -m sample reads/length_filtered reads/dereplicated   # -m sample creates the MERGED_sample tag obi clean -s needs

# Remove singletons
# count >=2: removes sequencing errors; increase to 5-10 for noisy datasets
obi grep -p 'sequence["COUNT"] >= 2' reads/dereplicated reads/denoised   # obi uniq writes the COUNT tag uppercase (case-sensitive)

# Denoise (remove PCR/sequencing errors)
# ratio 0.05: sequences <5% abundance of a 1-mismatch parent are merged
obi clean -s MERGED_sample -r 0.05 -H reads/denoised reads/cleaned

Path B: DADA2 (R)

library(dada2)

path <- 'trimmed/'
fnFs <- sort(list.files(path, pattern = '_R1.fastq.gz', full.names = TRUE))
fnRs <- sort(list.files(path, pattern = '_R2.fastq.gz', full.names = TRUE))
sample_names <- gsub('_R1.fastq.gz', '', basename(fnFs))

# Filter and trim
# truncLen: set based on quality profiles; marker-dependent
# maxEE c(2,2): max expected errors; standard for eDNA
# minLen 50: minimum after trimming
filtFs <- file.path('filtered', paste0(sample_names, '_F_filt.fastq.gz'))
filtRs <- file.path('filtered', paste0(sample_names, '_R_filt.fastq.gz'))
out <- filterAndTrim(fnFs, filtFs, fnRs, filtRs,
                     truncLen = c(200, 180), maxEE = c(2, 2),
                     minLen = 50, truncQ = 2, rm.phix = TRUE,
                     multithread = TRUE)

# Learn error rates
errF <- learnErrors(filtFs, multithread = TRUE)
errR <- learnErrors(filtRs, multithread = TRUE)

# Denoise
dadaFs <- dada(filtFs, err = errF, multithread = TRUE)
dadaRs <- dada(filtRs, err = errR, multithread = TRUE)

# Merge paired reads
# minOverlap 20: standard; increase if amplicon has short overlap region
merged <- mergePairs(dadaFs, filtFs, dadaRs, filtRs, minOverlap = 20)

# Build ASV table
seqtab <- makeSequenceTable(merged)

# Remove chimeras
# method 'consensus': standard; 'pooled' for higher sensitivity
seqtab_nochim <- removeBimeraDenovo(seqtab, method = 'consensus', multithread = TRUE)
chimera_rate <- 1 - sum(seqtab_nochim) / sum(seqtab)
message(sprintf('Chimera rate: %.1f%%', chimera_rate * 100))

QC Checkpoint: Denoising

# Gate 1: Chimera rate <20%
if (chimera_rate > 0.20) message('WARNING: High chimera rate. Check primer removal and PCR conditions.')

# Gate 2: ASV count reasonable for marker
n_asvs <- ncol(seqtab_nochim)
message(sprintf('ASVs after denoising: %d', n_asvs))
# Typical ranges: COI 500-5000, 12S 50-500, ITS 200-3000

Step 4: Contamination Filtering

R (decontam)

library(decontam)
library(phyloseq)

ps <- phyloseq(otu_table(seqtab_nochim, taxa_are_rows = FALSE),
               sample_data(meta))

# Identify negative controls
sample_data(ps)$is_neg <- sample_data(ps)$sample_type == 'negative_control'

# Combined method (Davis 2018): uses BOTH negative controls AND DNA concentration
# threshold=0.1 default; 0.05 for low-biomass samples
# CRITICAL: output is SCREENING, not classification; biological-plausibility check required
if ('dna_concentration' %in% sample_variables(ps)) {
    contam <- isContaminant(ps, method = 'combined', neg = 'is_neg',
                            conc = 'dna_concentration', threshold = 0.1)
} else {
    # Fall back to prevalence-only if no qPCR/Qubit DNA-concentration data
    contam <- isContaminant(ps, method = 'prevalence', neg = 'is_neg',
                            threshold = 0.1)
}
message(sprintf('Flagged candidate contaminant ASVs: %d', sum(contam$contaminant)))
message('Manual review required: verify biological plausibility before deletion')

ps_clean <- prune_taxa(!contam$contaminant, ps)

# Remove negative control samples
ps_clean <- subset_samples(ps_clean, sample_type != 'negative_control')

Tag-jumping removal (Schnell 2015; NovaSeq caveat)

# Tag-jumping: cross-contamination from index hopping during library prep / sequencing
# Schnell 2015 Mol Ecol Resour 15:1289-1303 documented 0.1-2% per read pair
# NovaSeq patterned flow cells have ~10x higher rates than MiSeq
# Use platform-appropriate threshold:
#   MiSeq: 0.001-0.005 (0.1-0.5% of ASV total)
#   NovaSeq: 0.005-0.01 (0.5-1% of ASV total)
# Quantify residual rate first from per-ASV cross-sample appearance
otu <- as(otu_table(ps_clean), 'matrix')
max_per_asv <- apply(otu, 2, max)
otu_filtered <- otu
# Set threshold by platform; default below is for MiSeq
tag_jump_frac <- 0.001
for (j in 1:ncol(otu)) {
    threshold <- max_per_asv[j] * tag_jump_frac
    otu_filtered[otu[, j] < threshold, j] <- 0
}
otu_table(ps_clean) <- otu_table(otu_filtered, taxa_are_rows = FALSE)
# Modern alternative: metabaR::tagjumpslayer(metabarlist_obj, threshold = 0.03)

QC Checkpoint: Decontamination

# Gate: verify contaminant ASVs were removed from real samples
n_before <- ntaxa(ps)
n_after <- ntaxa(ps_clean)
message(sprintf('ASVs removed as contaminants: %d (%.1f%%)',
                n_before - n_after, (n_before - n_after) / n_before * 100))
if ((n_before - n_after) / n_before > 0.5) {
    message('WARNING: >50% ASVs flagged as contaminants. Review decontam threshold.')
}

Step 5: Taxonomy Assignment

OBITools3 (ecotag)

# ecotag assigns taxonomy using LCA algorithm against reference database
# Reference databases: EMBL, BOLD, MIDORI2, UNITE (marker-dependent)
obi ecotag -R reads/refdb --taxonomy reads/taxonomy reads/cleaned reads/assigned

# Filter by assignment quality (species-level for COI)
obi grep -p 'sequence["BEST_IDENTITY"] >= 0.97' reads/assigned reads/filtered_assigned   # ecotag writes BEST_IDENTITY uppercase (case-sensitive)

obi export --tab-output reads/filtered_assigned > taxonomy_results.tsv

DADA2 (assignTaxonomy)

# SILVA for 16S/18S, UNITE for ITS, custom for COI/12S
# Reference databases must be formatted for DADA2
# minBoot 50: minimum bootstrap confidence; 80 for more conservative assignments
taxa <- assignTaxonomy(seqtab_nochim, 'reference_db.fa.gz', multithread = TRUE, minBoot = 50)
taxa <- addSpecies(taxa, 'species_db.fa.gz')

Marker-specific taxonomy databases

QC Checkpoint: Taxonomy

# Gate: assignment rate should meet marker expectations
assigned <- !is.na(taxa[, 'Phylum'])
assignment_rate <- sum(assigned) / length(assigned) * 100
message(sprintf('Taxonomy assignment rate (phylum level): %.1f%%', assignment_rate))
if (assignment_rate < 60) message('WARNING: Low assignment rate. Check reference database completeness.')

Step 6: Diversity Analysis

R (iNEXT)

library(iNEXT)

otu_matrix <- as(otu_table(ps_clean), 'matrix')

# Hill numbers: q=0 (richness), q=1 (Shannon diversity), q=2 (Simpson diversity).
# Default endpoint = per-sample 2x reference size (the Chao 2014 doubling rule); do NOT hardcode a
# single global 2*max, which over-extrapolates small samples under unequal library sizes.
inext_out <- iNEXT(as.list(as.data.frame(t(otu_matrix))),
                   q = c(0, 1, 2), datatype = 'abundance')

# Coverage-based standardization at C=0.95: the fair cross-sample comparison (equalizes completeness,
# not raw depth). estimateD returns Hill numbers at that coverage.
div_c95 <- estimateD(as.list(as.data.frame(t(otu_matrix))),
                     q = c(0, 1, 2), datatype = 'abundance', base = 'coverage', level = 0.95)

# Sample completeness diagnostic: fraction of estimated diversity observed
completeness <- inext_out$DataInfo$SC
message(sprintf('Sample completeness range: %.1f%% - %.1f%%',
                min(completeness) * 100, max(completeness) * 100))

QC Checkpoint: Diversity

# Gate 1: rarefaction approaching asymptote
if (min(completeness) < 0.80) {
    message('WARNING: Some samples have low completeness (<80%). Deeper sequencing recommended.')
}

# Gate 2: richness. Use the COVERAGE-STANDARDIZED q=0, NOT inext_out$AsyEst 'Species richness' (the
# Chao1 asymptotic): denoising stripped singletons (COUNT >= 2), leaving Chao1 no f1, so it degenerates
# to observed richness -- the collapse rule 3 forbids. estimateD returns Order.q / qD.
q0_c95 <- div_c95[div_c95$Order.q == 0, ]
message(sprintf('Coverage-standardized (C=0.95) richness range: %.0f - %.0f effective species',   # qD is fractional; %d errors on doubles
                min(q0_c95$qD), max(q0_c95$qD)))

Step 7: Community Comparison

R (vegan + indicspecies)

library(vegan)
library(indicspecies)

otu_matrix <- as(otu_table(ps_clean), 'matrix')
env_data <- as(sample_data(ps_clean), 'data.frame')

# Hellinger transformation: standard for community composition data
# Reduces influence of dominant species
otu_hell <- decostand(otu_matrix, method = 'hellinger')

# DCA on untransformed data to determine gradient length
dca <- decorana(otu_matrix)
gradient_length <- diff(range(scores(dca, display = 'sites', choices = 1)))
message(sprintf('DCA gradient length: %.2f SD', gradient_length))

# RDA: linear response (<=3 SD), uses Hellinger-transformed data
# CCA: unimodal response (>3 SD), uses raw abundances (chi-squared distance)
if (gradient_length <= 3) {
    ord <- rda(otu_hell ~ temperature + depth + season, data = env_data)
    method_name <- 'RDA'
} else {
    ord <- cca(otu_matrix ~ temperature + depth + season, data = env_data)
    method_name <- 'CCA'
}

# Permutation test for significance
# permutations 999: standard; increase to 9999 for publication
anova_result <- anova.cca(ord, permutations = 999)
message(sprintf('%s significance: p = %.4f', method_name, anova_result$`Pr(>F)`[1]))

# MANDATORY companion: PERMANOVA + PERMDISP (Anderson & Walsh 2013 Ecol Monogr 83:557-574)
# adonis2 tests centroid difference; betadisper tests dispersion homogeneity
# If betadisper is also significant, PERMANOVA significance is dispersion-confounded
bray_dist <- vegdist(otu_matrix, method = 'bray')
permanova <- adonis2(bray_dist ~ site, data = env_data,
                     by = 'margin', permutations = 999)
disp <- betadisper(bray_dist, env_data$site)
disp_test <- permutest(disp, permutations = 999)
message(sprintf('PERMANOVA p = %.4f; PERMDISP p = %.4f',
                permanova[['Pr(>F)']][1], disp_test$tab[['Pr(>F)']][1]))
if (permanova[['Pr(>F)']][1] < 0.05 && disp_test$tab[['Pr(>F)']][1] < 0.05) {
    message('WARNING: Both PERMANOVA and PERMDISP significant; location-vs-dispersion confounded')
}

# Indicator species analysis with group-size equalization (NOT basic IndVal)
# func='IndVal.g' corrects for unbalanced group sizes (De Caceres & Legendre 2009)
indval <- multipatt(otu_matrix, env_data$site, func = 'IndVal.g',
                    control = how(nperm = 999))
summary(indval, alpha = 0.05)

Parameter Recommendations

Common Errors

Related Skills

• ecological-genomics/edna-metabarcoding - Detailed eDNA processing
• ecological-genomics/biodiversity-metrics - Diversity analysis details
• ecological-genomics/community-ecology - Ordination and indicator species
• read-qc/quality-reports - Raw read quality assessment
• reporting/automated-qc-reports - Aggregate FastQC across samples with MultiQC (a triage snapshot, not a pass/fail gate)
• microbiome/amplicon-processing - 16S clinical alternative

References

• Lamb PD, Hunter E, Pinnegar JK, et al (2019) How quantitative is metabarcoding? A meta-analytical approach. Molecular Ecology 28:420-430. DOI 10.1111/mec.14920. (read counts are not abundance.)
• Schnell IB, Bohmann K, Gilbert MTP (2015) Tag jumps illuminated - reducing sequence-to-sample misidentifications in metabarcoding studies. Molecular Ecology Resources 15:1289-1303. DOI 10.1111/1755-0998.12402. (tag jumping.)
• Davis NM, Proctor DM, Holmes SP, Relman DA, Callahan BJ (2018) Simple statistical identification and removal of contaminant sequences in marker-gene and metagenomics data. Microbiome 6:226. DOI 10.1186/s40168-018-0605-2. (decontam.)
• Anderson MJ, Walsh DCI (2013) PERMANOVA, ANOSIM, and the Mantel test in the face of heterogeneous dispersions. Ecological Monographs 83:557-574. DOI 10.1890/12-2010.1. (PERMANOVA + PERMDISP.)