FASTA and FASTQ parsing and writing in Rust.

This library provides an(other) attempt at parsing of the sequence formats FASTA and FASTQ, as well as writing.

Features:

• Fast parsers
• Informative errors
• Support for seeking
• Flexible methods for writing FASTA and FASTQ
• Serde support (for owned data structures)
• Functions for parallel processing

The FASTA parser can read and write multi-line files and allows to iterate over the sequence lines without doing any allocation or copying. The FASTQ parser does not support multiple sequence / quality lines.

Simple example: Reads FASTA sequences from STDIN and writes long enough ones to STDOUT, otherwise prints a message. ```rust use seq_io::fasta::{Reader,Record}; use std::io;

let mut reader = Reader::new(io::stdin()); let mut stdout = io::stdout();

while let Some(result) = reader.next() { let record = result.unwrap(); // determine sequence length without having to allocate // the whole sequence let seqlen = record.seqlines() .fold(0, |l, seq| l + seq.len()); if seqlen > 100 { record.writewrap(&mut stdout, 80).unwrap(); } else { println!("{} is only {} long", record.id().unwrap(), seqlen); } } ```

Records are directly borrowing data from the internal buffered reader, unless to_owned() is called, which creates an owned Record. This is also done by the the reader from the fastq-rs crate. By default, the buffer will automatically grow if a record is too large to fit in. How it grows can be configured, it is also possible to set a size limit.

Note: Make sure to add lto = true to the release profile in Cargo.toml because calls to functions of the underlying buffered reader (buf_redux) are not inlined otherwise.

View documentation

Multi-threaded processing

The parallel module contains functions for sending FASTQ/FASTA records to a thread pool where expensive calculations are done. Sequences are processed in batches (RecordSet) because sending across channels has a performance impact. FASTA/FASTQ records can be accessed in both the 'worker' function and (after processing) a function running in the main thread.

Performance comparisons

The FASTQ reader from this crate performs similar to the fastq-rs reader. The rust-bio readers are slower due to allocations, copying, and UTF-8 validity checks.

All comparisons were run on a set of 100,000 auto-generated, synthetic sequences of uniform length (500 bp) loaded into memory. The parsers from this crate (seqio) are compared with fastq-rs (fastqrs) and Rust-Bio (bio). The bars represent the throughput in GB/s, the error bars show the +/- deviation as inferred from the deviations provided by cargo bench, that is: (max_time - min_time) / 2 used per iteration. Run on a Mac Pro (Mid 2010, 2.8 GHz Quad-Core Intel Xeon, OS X 10.13) using Rust 1.23 nightly

FASTA

FASTQ

Explanation of labels:

• Top bars: Iteration over all records without further action.
• owned: An owned copy of each record is created for comparison with Rust-Bio, which does not provide zero copy parsing.
• multiline: The FASTA sequence is split into 5 x 100 bp lines.
• recset: Records are parsed into record sets using read_record_set() (involves some copying).
• parallel: Record sets are are sent to worker threads for parallel processing where they are being iterated over and then sent back to the main thread where there is another iteration over the records (the latter only in seq_io)