search

ComparativeGenomicsToolkit / hal

Hierarchical Alignment Format
Other
165 stars 39 forks source link

readme

[Build Status]

Hierarchical Alignment (HAL) Format API (v2.3)

Copyright (C) 2012 - 2014 by Glenn Hickey (hickey@soe.ucsc.edu) Copyright (C) 2012-2023 by UCSC Computational Genomics Lab Released under the MIT license, see LICENSE.txt

HAL is a structure to efficiently store and index multiple genome alignments and ancestral reconstructions. HAL is a graph-based representation which provides several advantages over matrix/block-based formats such as MAF, such as improved scalability and the ability to perform queries with respect to an arbitrary reference or subtree.

This package includes the HAL API and several analysis and conversion tools which are described below. HAL files are presently stored in either HDF5 or mmap format, but we note that the tools and most of the API are format-independent, so other databases could be implemented in the future.

Citing

Glenn Hickey, Benedict Paten, Dent Earl, Daniel Zerbino, and David Haussler. HAL: A Hierarchical Format for Storing and Analyzing Multiple Genome Alignments. Bioinformatics. 2013. Advance Online Access

Table of Contents

Code Contributors

Installation

Note HAL is included in Cactus, which is released as pre-built binaries and Docker images

Requirements

Downloading HAL

From the parent directory of where you want HAL installed:

 git clone https://github.com/ComparativeGenomicsToolkit/hal.git

Installing Dependencies

HDF5 1.10.1 with C++ API enabled

sonLib

From the same parent directory where you downloaded HAL:

  git clone https://github.com/ComparativeGenomicsToolkit/sonLib.git
  pushd sonLib && make && popd

If sonLib and HAL are not sister directories, set the sonLibRootDir make variable to reflect the absolute pathname of the directory where you installed sonLib:

sonLibRootDir=/path/to/sonLib

either in an include.local.mk file top level hal/ directory, or as a make command argument, e.g.:

pushd sonLib && make sonLibRootDir=/path/to/sonLib && popd

Optional support of reading HAL files over HTTP via UCSC's URL Data Cache (UDC)

Define ENABLE_UDC before making, and specify the path of the Kent source tree using KENTSRC. When built with this enabled, all HAL files opened read-only will be accessed using UDC which supports both local files and URLs.

  export  ENABLE_UDC=1
  export  KENTSRC=<path to top level of Kent source tree>

Those without the UCSC genome browser already installed locally will probably find it simpler to first mount URLs with HTTPFS before opening with HAL.

Optional support of PhyloP evolutionary constraint annotation

PhyloP is part of the Phast Package, and can be used to test for genomic positions that are under selective pressure. We are working on prototype support for running PhyloP on HAL files. In order to enable this support, Phast must be installed. We recommend downloading the latest source using Subversion.

From the same parent directory where you downloaded HAL:

Special thanks to Melissa Jane Hubiz and Adam Siepel from Cornell University for their work on extending their tools to work with HAL.

Building HAL

From the hal/ directory:

  make

Before using HAL, add it to your path:

  export PATH=<path to hal>/bin:${PATH}

The parent directory of hal/ should be in your PYTHONPATH in order to use any of the Python functionality. This includes running make test

  export PYTHONPATH=<parent of hal>:${PYTHONPATH}

HAL Tools

General Options

Detailed command line options can be obtained by running each tool with the --help option.

Two stored formats are included with HAL: HDF5 and mmap. HDF5 is standard container format for larger data sets with good compression characteristics . The mmap format stores the raw data structures in a file, which is access by mapping in into memory using the mmap system call. HAL files in the mmap format a considerably bigger but often much faster to access. The halExtract command can be used to copy between formats.

All HAL tools compiled with HDF5 support expose some caching parameters. Tools that create HAL files also include chunking and compression parameters. In most cases, the default values of these options will suffice.

--cacheBytes <value>: The maximum size of each array cache. 3 such caches can be allocated per genome in the alignment.

--cacheRDC <value>: The number of slots in each cache. This number should be set to a prime number that is roughly 50 x [cacheBytes / chunk].

--cacheMDC <value>: Size of the metadata cache. There is presently no reason to touch this.

--chunk <value>: The chunk size for the hdf5 arrays. Unreasonable chunk sizes can adversely affect cache performance. Larger chunks can lead to better compression. [default = 1000]

--deflate <value>: Compression level. Higher levels tend to not significantly decrease file sizes but do increase run time. [0:none - 9:max] [default = 2]

--inMemory: Load all data in memory (and disable hdf5 cache). [default = False]

Importing from other formats

MAF Import

MAF is a text format used at UCSC to store genome alignments. MAFs are typically stored with respect to a reference genome. MAFs can be imported into HAL as subtrees using the maf2hal command.

To import primates.maf as a star tree where the first alignment row specifies the root, and all others the leaves:

 maf2hal primates.maf primates.hal

To import primates.maf using "chimp" has the root

 maf2hal primates.maf primates.hal --refGenome chimp

The more the underlying tree looks like a star tree, the less efficient HAL is as all genomes will be fragmented with respect to each other. If ancestral (or multiple reference) sequences are available, or if it is acceptable to use a non-reference species as a reference proxy, then trees of arbitrary typologies can be constructed using the --append option.

 maf2hal mammals.maf mammals.hal --refGenome mouse --targetGenomes human,rat,chimp,dog
  maf2hal mammals.maf mammals.hal --append --refGenome human --targetGenomes chimp,gorilla,orang
  maf2hal mammals.maf mammals.hal --append --refGenome dog --targetGenomes cow,horse

This will create a tree that looks like

 ((chimp, gorilla,orang)human, rat,(cow,horse)dog)mouse;

Cactus Import

HAL is most beneficial when consensus reference or ancestral sequences are available at the internal nodes of the tree. This is the type of information generated by progressive alignment pipelines. Cactus is our implementation of such a pipeline.

Exporting to other formats

MAF Export

Please see the Maf Export / cactus-hal2maf documentation in the Progressive Cactus Manual. hal2mafMP.py is deprecated and we strongly recommend against running hal2maf directly. Use cactus-hal2maf instead.

FASTA Export

DNA sequences (without any alignment information) can be extracted from HAL files in FASTA format using hal2fasta.

Pangenome Graph Export (GFA and VG)

A HAL file can be converted into a pangenome using hal2vg, which can be downloaded as a standalone binary here.

It is also possible to go through PAF and seqwish as follows. But beware: This approach will currently miss alignments between snps:

hal2fasta mammals.hal $(halStats --root mammals.hal) --subtree --upper --ucscSequenceNames > mammals.fa
hal2paf mammals.hal --inMemory > mammals.paf
seqwish -p mammals.paf -s mammals.fa -g mammals.gfa

# Note, in the above, genome names are prepended to fasta sequence names when exporting paths to the graph.
# To turn this off, remove --ucscSequenceNames from hal2fasta and add --onlySequenceNames to hal2paf

This graph can then be imported into a compressed format to work with vg

vg convert -g mammals.gfa -p > mammals.pg

Displaying in the UCSC Genome Browser using Assembly Hubs

HAL alignments can be displayed as Assembly Hubs in the Genome Browser. To create an assembly hub, use the Comparative Annotation Toolkit or run

hal2assemblyHub.py ./jobStore mammals.hal outputDirectory

Larger alignments require the use of the --lod option to generate precomputed levels of detail. ./jobStore here is just a location for Toil to use for temporary files.

Note that this process is presently dependent on having UCSC's faToTwoBit installed. The outputDirectory must be accessible as a URL in order to load the hub. More details are available at hal2assemblyHub Manual.

Summary Information

halValidate

It is a good idea to check if a hal file is valid after creating it.

halValidate mammals.hal

halStats

Some global information from a HAL file can be quickly obtained using halStats. It will return the number of genomes, their phylogenetic tree, and the size of each array in each genome.

  halStats mammals.hal

The --tree, --sequences, and --genomes options can be used to print out only specific information to simplify iterating over the alignment in shell or Python scripts.

halSummarizeMtuations

A count of each type of mutation (Insertions, Deletions, Inversions, Duplications, Transpositions, Gap Insertions, Gap Deletions) in each branch of the alignment can be printed out in a table.

 halSummarizeMutations mammals.hal

Subtrees can be specified using the --targetGenomes or --rootGenome option. The --maxGap option is used to distinguish from small, 'gap' indels and larger indels. This distinction is somewhat arbitrary (but conventional). HAL allows gap indels to be nested within larger rearrangements: ex. an inversion with a gap deletion inside would be counted as a single inversion, but an inversion containing a non-gap event would be identified as multiple independent inversions.

 halSummarizeMutations mammals.hal --maxNFraction 0

will prevent rearrangements with missing data as being identified as such. More generally, if an insertion of length 50 contains c N-characters, it will be labeled as missing data (rather than an insertion) if c/N > maxNFraction.

Levels of Detail

Some applications such as genome browsers my need to quickly access high-level information about the alignment without scanning every segment. We provide tools to resample a HAL graph to compute a coarser-grained levels of detail to speed up subsequent analysis at different scales. To generate an output hal file based on a sampling of every 100 bases:

 halLodExtract mammals.hal mammals_100.hal 100

To generate a series of levels of details, such that each level of detail is 5x coarser than the previous, and that there are at most (approx.) 100 segments at the lowest level, use the following script:

 halLodInterpolate.py mammals.hal lod_summary.txt --scale 5 --maxBlock 100

Note that both tools have a --keepSequences option to specify whether or not the DNA sequences are stored in the output files.

Analysis

Liftover

Annotations in BED, ie tab-delimited files whose first three columns are

 SequenceName     StartPosition    LastPosition+1

can be lifted over between genomes using halLiftover. halLiftover does a base-by-base mapping between any two sequences in the alignment (following paralogy relations as well). The output is written in BED (default) or PSL format.

 halLiftover mammals.hal human human_annotation.bed dog dog_annotation.bed

will map all annotations in human_annotation.bed, which must refer to sequences in the human genome, to their corresponding locations in dog (if they exist), outputting the resulting annotations in dog_annotation.bed

halLiftover attempts to autodetect the BED version of the input. This can be overried with the --inVedVersion option. Columns that are not described in the official BED specs can be optionally mapped as-is using the --keepExtra option.

By default, halLiftover uses spaces and/or tabs to separate columns. To use only tabs (ie to allow spaces within names), use the --tab option.

Annotations in Wiggle format can likewise be mapped using halWiggleLiftover

See also the Comparative Annotation Toolkit for generating and working with HAL annotations.

halSynteny

halSynteny applies some additional chaining after halLiftover to give syntenic blocks as described in the halSynteny paper.

Alignment Depth

The number of distinct genomes different bases of a set of target genomes align to can be computed using the halAlignmentDepth tool. The output is in .wig format.

Mutation Annotation

SNPs

To compute the point mutations (SNPs) between a given pair of genomes in the HAL graph, halSnps can be used:

 halSnps mammals.hal human duck --bed human_duck_snps.bed

will produce a BED files listing the SNPs in human coordinates between human and duck. A count of the number of snps and the total aligned columns are printed to stdout.

General mutations along branches

Annotation files, as described above, can be generated from the alignment to provide the locations of substitutions and rearrangements. Annotations are done on a branch-by-branch basis, but can be mapped back to arbitrary references using halLiftover if so desired. The produced annotation files have the format

 SequenceName     StartPosition    LastPosition+1  MutationID  ParentGenome ChildGenome

The ID's refer to the types of mutations described above, and are explained in the header of each generated file. To generate tables of rearrangement mutations between human and its most recent ancestor in the alignment, run

 halBranchMutations mammals.hal human --refFile ins.bed --parentFile del.bed

Two bed files must be specified because the coordinates of inserted (and by convention inverted and transposed) segments are with respect to bases in the human genome (reference), where as deleted bases are in ancestral coordinates (parent).

Point mutations can optionally be written using the --snpFile <file> option. The '--maxGap' and '--maxNFraction' options can specify the gap indel threshold and missing data threshold, respectively, as described above in the halSummarizeMtuations section.

Constrained Element Prediction

(Under development)

PhyloP is part of the Phast Package, and can be used to test for genomic positions that are under selective pressure. We are working on prototype support for running PhyloP on HAL files.

Special thanks to Melissa Jane Hubiz and Adam Siepel from Cornell University for their work on extending their tools to work with HAL.

Related Tools

These are links to other tools that can be used with HAL:

Cactus

Cactus is the alignment software for generating hal alignments for both progressive and pangenome alignments. It includes cactus-hal2maf, which is the recommended way of exporting to MAF. Cactus releases are also the easiest way to install HAL.

hal2vg

hal2vg is a converter for pangenome alignments. Note that it does not work reliably for large progressive alignments.

(hal2vg citation)

HALPER

HALPER is a tool for making continuous orthologs from outputs of halLiftover.

(HALPER citation)

Example of HAL Genome Representation

The following is obtained by running h5ls -v -r (included with hdf5) on an ancestral genome, in this case a small simulated human-chimp ancestor named sHuman-sChimp. The genome itself is stored as a group. It contains four important 1-dimensional arrays:

More information can be found in the manuscript:

Glenn Hickey, Benedict Paten, Dent Earl, Daniel Zerbino, and David Haussler. HAL: A Hierarchical Format for Storing and Analyzing Multiple Genome Alignments. Bioinformatics, Volume 29, Issue 10, 15 May 2013, Pages 1341–1342, https://doi.org/10.1093/bioinformatics/btt128