OrfM

A simple and not slow open reading frame (ORF) caller. No bells or whistles like frameshift detection, just a straightforward goal of returning a FASTA file of open reading frames over a certain length from a FASTA/Q file of nucleotide sequences.

As of version 2.0, it is a pure-Rust reimplementation of the original C OrfM. The algorithm is the same as the original, but the codebase has been rewritten from scratch in Rust, with a library API added.

Install

OrfM can be installed in different ways:

1) Install via shell pipe

Follow the instructions on the releases page e.g. for Linux:

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/wwood/OrfM/releases/download/v<version>/orfm-installer.sh | sh

2) Install from pre-compiled binaries

OrfM can be installed by downloading pre-compiled binaries available at https://github.com/wwood/OrfM/releases. Once you have downloaded the package, extract and run it e.g. for GNU/Linux:

tar xzf orfm-<version>.tar.gz
cd orfm-<version>
./orfm -h

3) Install from source (requires Rust toolchain)

cargo install orfm

Or build without installing:

cargo build --release
# binary at target/release/orfm

4) Install via Conda / Pixi

conda install -c bioconda orfm

5) Install with brew

Thanks to Torsten Seemann (@tseemann), OrfM can be installed through homebrew:

brew install brewsci/bio/orfm

Usage

orfm [OPTIONS] [INPUT]

Reads from stdin if no input file is given. Accepts FASTA or FASTQ, gzipped or uncompressed.

Options

Examples

# Basic usage
orfm input.fasta > orfs.faa

# From gzipped FASTQ, shorter minimum ORF length
orfm -m 30 reads.fastq.gz > orfs.faa

# Pipe from stdin, write transcripts
cat input.fasta | orfm -m 60 -t transcripts.fna > orfs.faa

# Use mitochondrial codon table
orfm -c 2 mito.fasta > orfs.faa

Output

The output ORFs fasta file contains any stretch of continuous codons which does not include a stop codon. There is no requirement for a start codon to be included in the ORF. One could say that OrfM is an ORF caller, not a gene caller (like say prodigal or genscan).

The output ORFs are named in a straitforward manner. The name of the sequence (i.e. anything before a space) is followed by _startPosition_frameNumber_orfNumber and then the comment of the sequence (i.e. anything after the space) is given after a space, if one exists. For example,

$ cat eg.fasta
>abc|123|name some comment
ATGTTA
$ orfm -m 3 eg.fasta
>abc|123|name_1_1_1 some comment
ML

The startPosition of reverse frames is the left-most position in the original sequence, not the codon where the ORF starts.

Library usage

orfm can be used as a Rust library. Add to your Cargo.toml:

[dependencies]
orfm = '*'

Then to use it:

use orfm::OrfCaller;

let caller = OrfCaller::new(1, 96, None).unwrap(); // table_id, min_length, position_limit

// Iterate over ORFs from a file
for orf in caller.call_from_file("input.fasta") {
    println!(">{}", orf.name());
    println!("{}", std::str::from_utf8(&orf.protein).unwrap());
}

// Or call on a single sequence
let orfs = caller.find_orfs("seq1", "", b"ATGGATGCTGAA...");
for orf in &orfs {
    let transcript = orf.transcript(b"ATGGATGCTGAA...");
    // ...
}

Benchmarking

Walltime (seconds) results on a Linux x86-64 server, on 1 million random 150 bp sequences:

┌─────────────────┬───────────────┬──────────────────┬────────┬┬────────────────────────┐
│      Input      │ OrfM v1.4 (C) │ OrfM v2.1 (Rust) │ getorf ││ OrfM (Rust) / OrfM (C) │
├─────────────────┼───────────────┼──────────────────┼────────┼┼────────────────────────┤
│ fasta_unwrapped │          2.14 │             1.67 │   8.77 ││                  0.78x │
├─────────────────┼───────────────┼──────────────────┼────────┼┼────────────────────────┤
│ fasta_wrapped   │          2.14 │             1.72 │   8.86 ││                  0.81x │
├─────────────────┼───────────────┼──────────────────┼────────┼┼────────────────────────┤
│ fasta_gzipped   │          2.57 │             1.92 │    N/A ││                  0.74x │
├─────────────────┼───────────────┼──────────────────┼────────┼┼────────────────────────┤
│ fastq           │          2.28 │             1.71 │   9.05 ││                  0.75x │
├─────────────────┼───────────────┼──────────────────┼────────┼┼────────────────────────┤
│ fastq_gzipped   │          2.75 │             1.67 │    N/A ││                  0.61x │
└─────────────────┴───────────────┴──────────────────┴────────┴┴────────────────────────┘

• A ratio < 1 means OrfM v2.1 (Rust) is faster than OrfM v1.4 (C); Rust is 19–39% faster depending on input type.
• getorf (EMBOSS) does not support gzipped input (N/A). On plain FASTA/FASTQ it is ~4–5× slower than OrfM v2.1 (Rust).
• Peak RSS memory usage is similar across all programs (~85 MB).
• All replicates produce identical output (verified by diff).

A Snakefile is included for comparing performance against the original C OrfM and getorf (EMBOSS). It generates 1 million random 150 bp sequences in various formats, runs all three tools (3 replicates), checks output correctness, and collects wall-clock time and peak RSS memory.

pixi run snakemake -j1
cat benchmark/results.tsv
cat benchmark/correctness.txt

Requires the C OrfM binary at ~/git/OrfM/orfm (path configurable in the Snakefile). getorf is installed automatically via the pixi environment (emboss package).

Notes

• Exposes a Rust library API with an iterator over translated ORFs
• Supports all codon tables that OrfM supports (NCBI tables 1–25)
• Uses needletail for sequence parsing and aho-corasick for stop codon detection

License

OrfM is licensed under the GNU Lesser General Public License v3.0 (LGPL-3.0).

Citation

Since the algorithm was devised for the original version of OrfM, best to cite that:

Ben J. Woodcroft, Joel A. Boyd, and Gene W. Tyson. OrfM: A fast open reading frame predictor for metagenomic data. (2016). Bioinformatics. doi:10.1093/bioinformatics/btw241.