Cutadapt

Cutadapt

Overview

Cutadapt finds and removes adapter sequences, primers, poly-A tails, and other unwanted sequence from sequencing reads in an error-tolerant way (configurable mismatch/indel rate), and can filter, trim by quality/length, or demultiplex reads at the same time. It differs from heuristic auto-detecting trimmers (e.g. fastp) by requiring you to specify the adapter/primer sequence(s) — which makes it the right tool whenever the adapter is known and exact matching matters: Nextera adapters in ATAC-seq, 3' adapters in small-RNA-seq, PCR primers in amplicon sequencing, or barcodes for demultiplexing.

Installation

uv pip install cutadapt
# or
conda install -c bioconda cutadapt

Check version: cutadapt --version (targets the 5.x series; flag behavior below matches 5.0+ — see pitfalls for what changed at the 5.0 major bump).

When to Use vs. fastp

Use both in the same pipeline if useful: fastp for general QC/deduplication, cutadapt for precise, sequence-specific adapter/primer removal.

Adapter Types

• -a ADAPTER — 3' adapter (trims the adapter and everything after it).
• -g ADAPTER — 5' adapter (trims the adapter and everything before it).
• -b ADAPTER — either end (searches both).
• Anchored: -g ^ADAPTER (must be at the very start) / -a ADAPTER$ (must be at the very end) — much faster and less error-prone for known-position primers/barcodes.
• Linked adapters (both ends of a fragment, e.g., amplicon primers): -g ^PRIMER1...PRIMER2$ or via --pair-adapters.
• IUPAC wildcards are supported in adapter sequences (N, R, Y, etc.).
• Multiple adapters: repeat -a/-g — cutadapt tries all and reports which one matched (used for demultiplexing).

Quick Start: Single-End

cutadapt \
  -a AGATCGGAAGAGC \
  -o trimmed.fastq.gz \
  input.fastq.gz \
  > report.txt

Paired-End

cutadapt \
  -a AGATCGGAAGAGC -A AGATCGGAAGAGC \
  -o trimmed_R1.fastq.gz -p trimmed_R2.fastq.gz \
  input_R1.fastq.gz input_R2.fastq.gz

Lowercase flags (-a, -g, -b, -q) apply to R1; uppercase (-A, -G, -B, -Q) apply to R2. Cutadapt keeps R1/R2 synchronized automatically — never trim the two files independently with separate commands, or read order/pairing breaks.

Quality, Length, and N Filtering

# -q 20,20    trim low-quality bases from both ends (Phred cutoff)
# -m 20       discard reads shorter than 20nt after trimming
# --max-n 0.1 discard reads with too many Ns
cutadapt \
  -a AGATCGGAAGAGC \
  -q 20,20 \
  -m 20 \
  --max-n 0.1 \
  -o trimmed.fastq.gz input.fastq.gz

Common companions: --discard-untrimmed (keep only reads where the adapter was actually found — useful for amplicon primer confirmation), --trim-n (strip trailing Ns), -u 10 (unconditionally cut 10 bases from the 5' end, e.g. a fixed UMI/spacer).

Demultiplexing

cutadapt \
  -g file:barcodes.fasta \
  -o "trimmed-{name}.fastq.gz" \
  input.fastq.gz

barcodes.fasta holds one named barcode sequence per record; cutadapt writes a separate output file per matched barcode ({name} is substituted), plus an unknown bucket for unmatched reads.

Small RNA / Poly-A Specifics

# Small RNA: keep short reads, trim the 3' adapter, discard anything too short to be biological signal
cutadapt -a TGGAATTCTCGGGTGCCAAGG -m 18 -M 30 -o trimmed.fastq.gz input.fastq.gz

# Poly-A tail removal (RNA-seq) — --poly-a supersedes the older `-a "A{100}"` recipe
cutadapt --poly-a -o trimmed.fastq.gz input.fastq.gz

Multi-Core and Reports

cutadapt -j 8 -a AGATCGGAAGAGC -o trimmed.fastq.gz input.fastq.gz   # -j 0 = auto-detect cores
cutadapt --json=report.json -a AGATCGGAAGAGC -o trimmed.fastq.gz input.fastq.gz

--json reports (adapter match counts, length histograms) feed cleanly into MultiQC. --info-file=info.txt (and --info-file-paired for R2) writes one line per read documenting exactly what was trimmed — useful for debugging unexpected trimming behavior.

Common Pitfalls

• Version 5.0 changed default compression and demux behavior. Since 5.0, default gzip compression level dropped from 5 to 1 (faster, larger intermediate files — fine since they're usually deleted downstream; use --compression-level=5 to match pre-5.0 output size). Also since 5.0, multi-adapter/barcode matches that are ambiguous between two adapters are no longer arbitrarily assigned to one — they go to unknown instead, which can silently increase your "unmatched" bucket size if you're comparing against older-cutadapt pipelines.
• Running R1/R2 through separate invocations. Always trim paired files together with -o/-p in one command; separate commands can desynchronize which reads survive filtering in each file.
• Not anchoring adapters that are always at a fixed position. Un-anchored search is slower and occasionally matches spuriously in the middle of a read; use ^/$ anchoring for primers/barcodes you know are at the read boundary.
• Confusing -a/-g orientation. -a trims the adapter and everything after; -g trims the adapter and everything before. Getting this backwards silently drops the biologically relevant part of the read instead of the adapter.
• Forgetting --discard-untrimmed for amplicon work. Without it, reads where the primer wasn't found are kept untrimmed rather than removed, which can leak primer-less off-target reads into downstream analysis.
• Python floor. cutadapt 5.2 dropped Python 3.9; pin cutadapt<5.2 if you must run on 3.9, or upgrade the environment.

Resources

• Docs: https://cutadapt.readthedocs.io/
• Changelog: https://cutadapt.readthedocs.io/en/stable/changes.html
• Source: https://github.com/marcelm/cutadapt
• Citation: Martin, M. (2011). Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet.journal, 17(1), 10-12. DOI: 10.14806/ej.17.1.200