16S/ITS Amplicon Analysis Pipeline

BioAnalysis.ca uses a distributed architecture with dedicated compute resources for each analysis job. No shared resources means zero cross-contamination risk and consistent performance.

Layer 1 — Upload-Time Validation (instant)

Runs during file upload, before any compute resources are used:

• Compression check — .gz files must start with the gzip signature 0x1f 0x8b; uncompressed .fastq files are also accepted and compressed on the analysis server
• FASTQ format check — decompresses first 4KB, verifies @header / sequence / + / quality 4-line format
• FASTA detection — rejects FASTA files (headers starting with >) with clear error
• Minimum size — rejects files <1KB (empty/corrupted)
• Sequence/quality length match — verifies line 2 and line 4 are equal length

Layer 2 — Pre-Pipeline Validation (~30 seconds)

Runs on the analysis server after download, before starting the pipeline:

• FASTQ integrity scan — validates 1,000 reads for correct 4-line format
• Read length distribution — flags reads <50bp as suspiciously short
• R1/R2 balance — paired-end files should have equal read counts (>80% ratio)
• Primer presence scan — checks 500 reads for expected primer sequences; warns if <5% match (wrong region or data type)
• Hard failure — pipeline aborts with descriptive error if no valid reads found

Layer 3 — Quality Assessment (post-analysis)

After analysis completes, every result is scored 0–100:

Verdict: Excellent (all pass) → Good (1 warning) → Fair (2+ warnings) → Poor (any fail)

NEW in v2.0 Instead of hardcoded truncation values, the system adapts to actual read quality:

Algorithm

1. Scan first 10,000 reads from R1 and R2 files
2. Compute median Phred quality score at each base position
3. Find the first position where median quality drops below Q25
4. Set truncation at that position (minimum floor: 150bp)
5. Safety caps: use 10th percentile read length (≥90% reads survive), never exceed region presets

Cutadapt v4.9

Removes primer sequences from the 5' end of reads. Reads without detectable primers are discarded.

Output: Trimmed FASTQ files + statistics (reads in/out, trim rate %)

DADA2 v1.30.0 (R/Bioconductor)

DADA2 infers exact Amplicon Sequence Variants (ASVs) from the quality-filtered reads. Unlike OTU clustering (97% similarity), ASVs resolve single-nucleotide differences.

Sub-steps

1. filterAndTrim() — truncate reads (quality-adaptive values), discard reads with >2 expected errors
2. learnErrors() — learn error model from the data (parametric error model)
3. dada() — denoise reads using the learned error model
4. mergePairs() — merge R1+R2 (paired-end), requiring ≥12bp overlap
5. makeSequenceTable() — create ASV × sample count matrix
6. removeBimeraDenovo() — remove chimeric sequences (consensus method)
7. assignTaxonomy() — classify ASVs using naive Bayesian classifier
8. addSpecies() — species-level assignment via exact matching

Output: ASV count table, representative sequences (FASTA), taxonomy assignments, denoising stats (reads at each stage)

SILVA v138.2 UNITE v9.0

Taxonomy is assigned at 7 ranks: Kingdom → Phylum → Class → Order → Family → Genus → Species. Species assignment requires 100% identity to a reference sequence.

MAFFT v7.520 FastTree v2.1.11

16S only — skipped for ITS (high indel rates make alignments unreliable).

1. MAFFT — multiple sequence alignment of all ASV representative sequences (FFT-NS-2 algorithm)
2. FastTree — approximate maximum-likelihood tree (GTR+CAT model)

Output: Newick tree file, used for Faith's Phylogenetic Diversity and UniFrac distances.

scikit-bio v0.6.2

PICRUSt2 v2.5.3 (16S only)

Predicts metagenomic functional content from 16S marker gene data:

1. Phylogenetic placement — places ASVs into a reference tree (EPA-ng)
2. Gene content prediction — predicts gene families using ancestral state reconstruction
3. Pathway inference — maps predicted genes to MetaCyc metabolic pathways (MinPath)

matplotlib v3.9 WeasyPrint v63.1 Gemini AI

Charts Generated

• Phylum composition — stacked bar chart of relative abundances
• Genus composition — top 20 genera bar chart
• Rarefaction curve — species accumulation curve (sampling depth vs. observed ASVs)

AI Interpretation

Each report section receives a contextual AI interpretation via Google Gemini (gemini-2.0-flash), providing clinical/ecological context for non-specialist users.

Deliverables

• Interactive Web Report — full results with charts, quality assessment, parameter transparency
• PDF Report — publication-ready document with all figures and tables
• Research Data ZIP — ASV table, taxonomy table, representative sequences, tree file, raw stats

Reference Databases