AdapterRemoval v3 - Dang fast(Q) processing
AdapterRemoval trims adapter sequences and low quality bases from High-Throughput Sequencing (HTS) data in FASTQ format, merges overlapping paired-end reads into (longer) consensus sequences, demultiplexes FASTQ reads containing 5' barcodes, and more.
AdapterRemoval v3 is a major revision of AdapterRemoval v2 that aims to simplify usage via sensible default settings. AdapterRemoval v3 expands the range of features available in v2 (see below), as well as greatly increasing throughput.
- Highlights
- Getting started
- Documentation
- Support
- Citation
Highlights
Features introduced in AdapterRemoval v3 are marked with a "(v3)":
- Supports Linux, MacOS, and Windows (v3)
- Trimming of adapter sequences from single-end and paired-end FASTQ reads
- Trimming of multiple, different adapters or adapter pairs
- Detailed human-readable and machine-readable QC reports (v3)
- The ability to perform QC-only runs with or without read processing (v3)
- Barcode-based demultiplexing with or without trimming of adapter sequences
- Support for samples identified by multiple barcode pairs (v3)
- Support for mixed orientation barcodes (v3)
- Support for multiple methods for trimming low quality bases/reads
- Quality trimming using windows or constant thresholds
- Quality trimming using the modified Mott algorithm (v3)
- Poly-X tail trimming, supporting any combination of trailing bases (v3)
- Enabled automatically for data generated using two-color sequencing technologies (e.g. Illumina iSeq, NovaSeq)
- Filtering of reads based on complexity (v3)
- Reconstruction of adapter sequences by pair-wise alignment of paired-end reads
- Merging of overlapping read-pairs into higher-quality consensus sequences
- Support for reading interleaved FASTQ files
- Support for flexible routing of output to files or STDOUT (v3)
- Support for writing FASTQ, SAM, and BAM files (v3)
- Support for SSE2, AVX2, AVX512, and NEON accelerated alignments (v3)
Performance
AdapterRemoval v3 features greatly increased throughput compared to AdapterRemoval v2. This is accomplished through support for additional SIMD instruction sets, and improved parallelization of I/O and computationally expensive tasks. To enable this, compression of output is now block-based, defaults to a lower compression level (4), and uses a more efficient implementation (libdeflate).
Please note that these results are preliminary:
Point labels indicate the number of threads configured on the command line. The X-axis shows actual CPU utilization. The Y-axis shows millions of 150bp reads processed per second in paired-end mode, with gzipped input and output.
Benchmarking was performed on an Intel i9-11900K with 8 physical cores, and plotting is therefore limited to ~8 CPUs. AdapterRemoval v2 was observed not to scale past 4 threads. AdapterRemoval v2 was run with default settings and --gzip-level 4 (gz4).
Getting started
Installation
Precompiled binaries are provided for Linux, MacOS, and Windows.