#bioinformatics #byte-range #file-processing #file-format #htsget #noodles #query

htsget-search

The primary mechanism by which htsget-rs interacts with, and processes bioinformatics files. It does this by using noodles to query files and their indices.

23 unstable releases (8 breaking)

0.9.1 Oct 17, 2024
0.8.1 Sep 6, 2024
0.7.1 May 22, 2024
0.6.6 Jan 2, 2024
0.1.4 Feb 24, 2023

#259 in Parser implementations

49 downloads per month
Used in 4 crates

MIT license

435KB
11K SLoC

htsget-search

MIT licensed Build Status

Creates URL tickets for htsget-rs by processing bioinformatics files. It:

  • Takes a htsget query and produces htsget URL tickets.
  • Uses noodles to process files.

Overview

This crate is the primary mechanism by which htsget-rs interacts with, and processes bioinformatics files. It does this by using noodles to query files and their indices. This crate contains abstractions that remove commonalities between file formats. Together with file format specific code, this defines an interface that handles the core logic of a htsget request. ht

Usage

For running htsget-rs as an application

This crate is responsible for handling bioinformatics file data. It supports BAM, CRAM, VCF and BCF files. For htsget-rs to function, files need to be organised in the following way:

  • Each file format is paired with an index. All files must have specific extensions.
    • BAM: File must end with .bam; paired with BAI index, which must end with .bam.bai.
    • CRAM: File must end with .cram; paired with CRAI index, which must end with .cram.crai.
    • VCF: File must end with .vcf.gz; paired with TBI index, which must end with .vcf.gz.tbi.
    • BCF: File must end with .bcf; paired with CSI index, which must end with .bcf.csi.
  • VCF files are assumed to be BGZF compressed.
  • BGZF compressed files (BAM, CRAM, VCF) can optionally also have a GZ index to make byte ranges smaller.

This is quite inflexible, and is likely to change in the future to allow arbitrary mappings of files and indices.

As a library

This crate has the following features:

  • The HtsGet trait represents an entity that can resolve queries according to the htsget spec. The htsget trait comes with a basic model to represent components needed to perform a search: Query, Format, Class, Tags, Headers, Url, Response. HtsGetFromStorage is the struct which is used to process requests.

Feature flags

This crate has the following features:

  • s3-storage: used to enable S3Storage functionality.
  • url-storage: used to enable UrlStorage functionality.
  • experimental: used to enable experimental features that aren't necessarily part of the htsget spec, such as Crypt4GH support through C4GHStorage.

Minimising Byte Ranges

One challenge involved with implementing htsget is meaningfully minimising the size of byte ranges returned in response tickets. Since htsget is used to reduce the amount of data a client needs to fetch by querying specific parts of a file, the data returned by htsget should ideally be as minimal as possible. This is done by reading the index file or the underlying target file, to determine the required byte ranges. However, this is complicated when considering BGZF compressed files.

For BGZF compressed files, htsget-rs needs to return compressed byte positions. Also, after concatenating data from URL tickets, the resulting file must be valid. This means that byte ranges must start and finish on BGZF blocks, otherwise the concatenation would not result in a valid file. However, index files (BAI, TBI, CSI) do not contain all the information required to produce minimal byte ranges. For example, consider this file:

  • There are 14 BGZF blocks positions using all available data in the corresponding index file (chunk start positions, chunk end positions, linear index positions, and metadata positions):
    • 4668, 256721, 499249, 555224, 627987, 824361, 977196, 1065952, 1350270, 1454565, 1590681, 1912645, 2060795 and 2112141.
  • Using just this data, the following query with:
    • referenceName=11, start=5015000, and end=5050000
  • Would produce these byte ranges:
    • bytes=0-4667
    • bytes=256721-1065951
  • However, an equally valid response, with smaller byte ranges is:
    • bytes=0-4667
    • bytes=256721-647345
    • bytes=824361-842100
    • bytes=977196-996014

To produce the smallest byte ranges, htsget-rs needs to find this data somewhere else. There are two ways to accomplish this:

  • Get the data from the underlying target file, by seeking to the start of a BGZF, and reading until the end of the block is found.
  • Get the data from an auxiliary index file, such as GZI.

Currently, htsget-rs takes the latter approach, and uses GZI files, which contain information on all BGZF start and end positions. However, this is not ideal, as GZI contains more information than required by htsget-rs. The former approach also has issues when considering cloud-based storage, which in the case of S3, does not have seek operations.

The way htsget-rs finds the information needed for minimal byte ranges is very likely to change in the future, as more efficient approaches are implemented. For example, a database could be used to further index files. Queries to a database could be as targeted as possible, retrieving only the required information.

Benchmarks

Since this crate is used to query file data, it is the most performance critical component of htsget-rs. Benchmarks, using Criterion.rs, are therefore written to test performance. Run benchmarks by executing:

cargo bench -p htsget-search --all-features

Alternatively if you are using cargo-criterion and want a machine readable JSON output, run:

cargo criterion --bench search-benchmarks --message-format=json -- LIGHT 1> search-benchmarks.json

License

This project is licensed under the MIT license.

Dependencies

~30–53MB
~1M SLoC