Crates.io | bio-streams |
lib.rs | bio-streams |
version | 0.5.0 |
source | src |
created_at | 2024-05-12 17:34:50.676893 |
updated_at | 2024-09-21 21:50:38.000785 |
description | Streaming bioinformatics data types |
homepage | |
repository | https://github.com/jeff-k/bio-streams |
max_upload_size | |
id | 1237677 |
size | 43,045 |
Webassembly examples: Remove non M. TB reads from streaming fastqs, amplicon based SARS-CoV-2 assembly
Shared Record
type by Fastq
and Fasta
streams:
pub struct Record<T: for<'a> TryFrom<&'a [u8]> = Vec<u8>> {
pub fields: Vec<u8>,
pub seq: T,
pub quality: Option<Vec<Phred>>, // fasta records set quality to `None`
}
Records can be read into custom types: pub struct Fastq<R: BufRead, T = Seq<Dna>>
// Open a pair of gzipped fastq files as streams of `Record`s with `Seq<Dna>` sequences
let fq1: Fastq<BufReader<MultiGzDecoder<File>>> = Fastq::new(BufReader::new(
MultiGzDecoder::new(File::open(&file1).unwrap()),
));
let fq2: Fastq<BufReader<MultiGzDecoder<File>>> = Fastq::new(BufReader::new(
MultiGzDecoder::new(File::open(&file2).unwrap()),
));
for zipped in fq1.zip(fq2) {
match zipped {
(Ok(r1), Ok(r2)) => {
// check that the last characters of the name strings are 1 and 2
if r1.fields[r1.fields.len() - 1] != b'1' || r2.fields[r2.fields.len() - 1] != b'2'
{
eprintln!("paired records do not end in 1/2");
}
// check that the description fields are equal up to the last character
if r1.fields[..r1.fields.len() - 1] != r2.fields[..r2.fields.len() - 1] {
eprintln!("reads do not have the same names");
}
}
_ => {
eprintln!("Parse error in fastq files");
}
}
}
To run the fqcheck
example program with read files r1.fq.gz
and f2.fq.gz
:
$ cargo build --example fqcheck --release
$ target/release/examples/fqcheck r1.fq.gz r2.fq.gz
// this opens a gzipped data stream and parses it into `Records` with `Seq<Amino>` sequence fields
let faa: Fasta<BufReader<File>, Seq<Amino>> =
Fasta::new(BufReader::new(File::open(&faa_file).unwrap()));
// we can convert amino acid k-mers directly into usizes and use them to index into a table
let mut histogram = Box::new([0u64; 1 << (K * Amino::BITS as usize)]);
for contig in faa {
// here "contig" is a fasta record
for kmer in contig.unwrap().seq.kmers::<K>() {
histogram[usize::from(kmer)] += 1;
}
}
To run the aminokmers
example program with fasta file proteins.faa
:
$ cargo build --example fqcheck --release
$ target/release/examples/aminokmers proteins.faa
input streams:
todo:
Phred
alias for u8