PacBio HiFi (single-end) workflow

PacBio HiFi 16S reads are long (~1.5 kb), single-end, and dominated by homopolymer indels rather than substitution errors. The workflow therefore differs from Illumina in a few key ways:

• Primers are removed, reads oriented, and quality/length filtered in one step — see step 1 below.
• No reverse reads and no merge step — denoised ASVs go straight into the table.
• --errfun pacbio is used instead of loess.
• The dada step uses PacBio-tuned alignment parameters: a wider band (--band 32), a relaxed homopolymer gap penalty (--homo-gap-p -1), and a larger k-mer screen (--kmer-size 7).

Do not use k=5 on HiFi reads

On ~1.5 kb reads the Illumina default --kmer-size 5 makes the pre-alignment screen a no-op (nearly every pair is fully aligned), so denoising runs several times slower. Use --kmer-size 7.

Keep learn and denoise params consistent

Pass the same --band, --homo-gap-p, and --kmer-size to both learn-errors and dada. If the error model is learned with different alignment params than denoising uses, dada will warn, and results change subtly. (See the param-mismatch note in Performance & Benchmarking.)

1. Remove primers, orient, and filter

HiFi reads come in both orientations and still carry the amplification primers. remove-primers (mirrors R's removePrimers()) trims them and, with --orient (on by default), flips reads that match only in the reverse-complement direction so every read points the same way — essential for PacBio, since the k-mer screen in later steps relies on consistently-oriented reads. It also accepts the same length/quality filters as filter-and-trim, so primer removal and filtering are done in a single pass (no separate filter-and-trim step needed):

dada2-rs remove-primers raw/sample.fastq.gz \
  --fout filtered/sample.fastq.gz \
  --primer-fwd AGRGTTYGATYMTGGCTCAG \
  --primer-rev RGYTACCTTGTTACGACTT \
  --trim-fwd --trim-rev --orient \
  --min-len 1000 --max-len 1600 \
  --max-n 0 --max-ee 2 --trunc-q 0 \
  --compress -o filtered/sample.json

The --primer-rev sequence is given 5′→3′ (catalog direction) and is reverse-complemented automatically before matching. Reads with no primer match are discarded.

Alternative: cutadapt

This step consolidates the removePrimers / filterAndTrim steps into one. You can also use a tool like cutadapt, which trims, filters, and reorients in one pass. We see slightly more reads passing filter with cutadapt at the same general parameters, likely due to more flexible primer matching.

2. Learn the error model

dada2-rs learn-errors filtered/*.fastq.gz \
  --nbases 100000000 --errfun pacbio --band 32 --homo-gap-p -1 --kmer-size 7 \
  -o errors_pacbio.json --verbose

3. Denoise each sample

dada2-rs dada filtered/sample.fastq.gz \
  --error-model errors_pacbio.json \
  --band 32 --homo-gap-p -1 --kmer-size 7 \
  -o dada/sample.json --verbose

Alternatively, store the PacBio parameters in the error-model JSON and pass --inherit-err-params so every dada call picks them up automatically.

4. Build sequence table and remove chimeras

There is no merge step — feed the per-sample dada JSON files directly into the table:

dada2-rs make-sequence-table dada/*.json -o seqtab.json
dada2-rs remove-bimera-denovo seqtab.json --method consensus -o seqtab_nochim.json