Skip to content
BranchesTags
 
 

Repository files navigation

wfmash

a pangenome-scale aligner

build and test install with bioconda DOI

wfmash is an aligner for pangenomes based on sparse homology mapping and wavefront inception.

wfmash uses a variant of MashMap to find large-scale sequence homologies. It then obtains base-level alignments using WFA, via the wflign hierarchical wavefront alignment algorithm.

wfmash is designed to make whole genome alignment easy. On a modest compute node, whole genome alignments of gigabase-scale genomes should take minutes to hours, depending on sequence divergence. It can handle high sequence divergence, with average nucleotide identity between input sequences as low as 70%.

wfmash is the key algorithm in pggb (the PanGenome Graph Builder), where it is applied to make an all-to-all alignment of input genomes that defines the base structure of the pangenome graph. It can scale to support the all-to-all alignment of hundreds of human genomes.

process

Each query sequence is broken into non-overlapping pieces defined by -s[N], --segment-length=[N]. These segments are then mapped using MashMap's mapping algorithm. Unlike MashMap, wfmash merges aggressively across large gaps, finding the best neighboring segment up to -c[N], --chain-gap=[N] base-pairs away.

Each mapping location is then used as a target for alignment using the wavefront inception algorithm in wflign. The resulting alignments always contain extended CIGARs in the cg:Z:* tag. Approximate mappings can be obtained with -m, --approx-map.

Sketching, mapping, and alignment are all run in parallel using a configurable number of threads. The number of threads must be set manually, using -t, and defaults to 1.

usage

wfmash has been developed to accelerate the alignment step in variation graph induction (the first step in the seqwish / smoothxg pipeline). Suitable default settings are provided for this purpose.

Seven parameters shape the length, number, identity, and alignment divergence of the resulting mappings.

mapping settings

These parameters affect the structure of the mappings:

  • -s[N], --segment-length=[N] is the length of the mapping seed (default: 1k). The best pairs of consecutive segment mappings are merged where separated by less than -c[N], --chain-gap=[N] bases.
  • -l[N], --block-length-min=[N] requires seed mappings in a merged mapping to sum to more than the given length (default 5kb).
  • -p[%], --map-pct-id=[%] is the percentage identity minimum in the mapping step
  • -n[N], --n-secondary=[N] is the maximum number of mappings (and alignments) to report for each segment above --block-length-min (the number of mappings for sequences shorter than the segment length is defined by -S[N], --n-short-secondary=[N], and defaults to 1)

By default, we obtain base-level alignments by applying a high-order version of WFA to the mappings. Various settings affect the behavior of the pairwise alignment, but in general the alignment parameters are adjusted based on expected divergence between the mapped subsequences. Specifying -m, --approx-map lets us stop before alignment and obtain the approximate mappings (akin to minimap2 without -c).

all-to-all mapping

Together, these settings allow us to precisely define an alignment space to consider. During all-to-all mapping, -X can additionally help us by removing self mappings from the reported set, and -Y extends this capability to prevent mapping between sequences with the same name prefix.

input indexing

wfmash requires a FASTA index (.fai) for its reference ("target"), and benefits if both reference and query are indexed. We can build these indexes on BGZIP-indexed files, which we recommend due to their significantly smaller size. To index your sequences, we suggest something like:

bgzip -@ 16 ref.fa
samtools faidx ref.fa.gz

Here, we apply bgzip (from htslib) to build a line-indexable gzip file, and then use samtools to generate the FASTA index, which is held in 2 files:

$ ls -l ref.fa.gz*
ref.fa.gz
ref.fa.gz.gzi
ref.fa.gz.fai

examples

Map a set of query sequences against a reference genome:

wfmash reference.fa query.fa >aln.paf

Setting a longer segment length forces the alignments to be more collinear:

wfmash -s 20k reference.fa query.fa >aln.paf

Self-mapping of sequences:

wfmash -X query.fa query.fa >aln.paf

Or just

wfmash query.fa >aln.paf

sequence indexing

wfmash provides a progress log that estimates time to completion.

This depends on determining the total query sequence length. To prevent lags when starting a mapping process, users should apply samtools index to index query and target FASTA sequences. The .fai indexes are then used to quickly compute the sum of query lengths.

Installation

Static binaries

We provide static builds of wfmash releases targeted at the x86-64-v3 instruction set.

Bioconda

wfmash recipes for Bioconda are available at https://anaconda.org/bioconda/wfmash. To install the latest version using Conda execute:

conda install -c bioconda wfmash

Building from Source

The build process for wfmash is managed using CMake, providing various options to customize the build.

Prerequisites

Before building wfmash, you need the following dependencies installed on your system:

  • GCC (version 9.3.0 or higher) or a recent version of Clang/LLVM
  • CMake
  • Zlib
  • GSL
  • HTSlib
  • LibLZMA
  • BZip2
  • Threads
  • OpenMP

On Ubuntu >20.04, these dependencies can be installed with the following command:

sudo apt install build-essential cmake zlib1g-dev libgsl-dev libhts-dev liblzma-dev libbz2-dev

Clone the Repository

Clone the wfmash repository:

git clone https://github.com/waveygang/wfmash.git
cd wfmash

Build Options

wfmash provides several CMake options to customize the build process:

  • BUILD_STATIC (default: OFF): Build a static binary.
  • BUILD_DEPS (default: OFF): Build external dependencies (htslib, gsl, libdeflate) from source. Use this if system libraries are not available or you want to use specific versions. HTSlib will be built without curl support, which removes a warning for static compilation related to dlopen.
  • BUILD_RETARGETABLE (default: OFF): Build a retargetable binary. When this option is enabled, the binary will not include machine-specific optimizations (-march=native).

These can be mixed and matched.

Building with System Libraries

To build wfmash using system libraries:

cmake -H. -Bbuild && cmake --build build -- -j 8

This command will configure and build wfmash in the build directory, using as many cores as you specify with the -j option.

Building with External Dependencies

If you need to build with external dependencies, use the BUILD_DEPS option:

cmake -H. -Bbuild -DBUILD_DEPS=ON && cmake --build build -- -j 8

This will download and build the necessary external dependencies.

Building a Static Binary

To build a static binary, use the BUILD_STATIC option:

cmake -H. -Bbuild -DBUILD_STATIC=ON && cmake --build build -- -j 16

Building a Retargetable Binary

To build a retargetable binary, use the BUILD_RETARGETABLE option:

cmake -H. -Bbuild -DBUILD_RETARGETABLE=ON && cmake --build build -- -j 8

This will configure the build without -march=native, allowing the binary to be run on different types of machines.

Installing

After building, you can install wfmash using:

cmake --install build

This will install the wfmash binary and any required libraries to the default installation directory (typically /usr/local/bin for binaries).

Tests

To build and run tests:

cmake --build build --target test

Notes for distribution

If you need to avoid machine-specific optimizations, use the CMAKE_BUILD_TYPE=Generic build type:

cmake -H. -Bbuild -D CMAKE_BUILD_TYPE=Generic && cmake --build build -- -j 8

The resulting binary should be compatible with all x86 processors.

Notes for debugging/plotting

To enable the functionality of emitting wavefront plots (in PNG format), tables (in TSV format), and timing information, add the -DWFA_PNG_TSV_TIMING=ON option:

cmake -H. -Bbuild -D CMAKE_BUILD_TYPE=Release -DWFA_PNG_TSV_TIMING=ON && cmake --build build -- -j 3

Note that this may make the tool a little bit slower.

nix

If you have nix, you can install directly from the repository via:

nix profile install github:waveygang/wfmash

For local development, from the wfmash repo directory:

nix build .#wfmash

And you can install into your profile from the source repo with:

nix profile install .#wfmash

guix

If you have guix:

guix build -f guix.scm

Docker and Singularity images with nix

Nix is also able to build an Docker image, which can then be loaded by Docker and converted to a Singularity image.

nix build .#dockerImage
docker load < result
singularity build wfmash.sif docker-daemon://wfmash-docker:latest

This can be run with Singularity like this:

singularity run wfmash.sif $ARGS

Where $ARGS are your typical command line arguments to wfmash.

Guix

installing via the guix-genomics git repository

First, clone the guix-genomics repository:

git clone https://github.com/ekg/guix-genomics

And install the wfmash package to your default GUIX environment:

GUIX_PACKAGE_PATH=. guix package -i wfmash

Now wfmash is available as a global binary installation.

installing via the guix-genomics channel

Add the following to your ~/.config/guix/channels.scm:

  (cons*
(channel
  (name 'guix-genomics)
  (url "https://github.com/ekg/guix-genomics.git")
  (branch "master"))
%default-channels)

First, pull all the packages, then install wfmash to your default GUIX environment:

guix pull
guix package -i wfmash

If you want to build an environment only consisting of the wfmash binary, you can do:

guix environment --ad-hoc wfmash

For more details about how to handle Guix channels, go to https://git.genenetwork.org/guix-bioinformatics/guix-bioinformatics.git.

running wfmash on a cluster

When aligning a large number of very large sequences, one wants to distribute the calculations across a whole cluster. This can be achieved by dividing the approximate mappings .paf into chunks of similar difficult alignment problems using split_approx_mappings_in_chunks.py.

example

  1. We restrict wfmash to its approximate mapping phase.
wfmash -m reference.fa query.fa > approximate_mappings.paf
  1. We use the Python script to split the approximate mappings into chunks. A good approximation of the number of chunks is the number of nodes on your cluster. In the following, we assume a cluster with 5 nodes.
python3 split_approx_mappings_in_chunks.py approximate_mappings.paf 5

This gives us:

ls
approximate_mappings.paf.chunk_0.paf
approximate_mappings.paf.chunk_1.paf
approximate_mappings.paf.chunk_2.paf
approximate_mappings.paf.chunk_3.paf
approximate_mappings.paf.chunk_4.paf
  1. Dependent on your cluster workload manager, create a command line to submit 5 jobs to your cluster.

One example without specifying a workflow manager:

wfmash -i approximate_mappings.paf.chunk_0.paf reference.fa query.fa > approximate_mappings.paf.chunk_0.paf.aln.paf

The resulting .paf can be directly plugged into seqwish.

# list all base-level alignment PAFs
PAFS=$(ls *.aln.paf | tr '\n' ',')
# trim of the last ','
PAFS=${PAFS::-1}
seqwish -s reference.fa -p $PAFS -g seqwish.gfa

use nf-core/pangenome

If you have Nextflow and Docker or Singularity available on your cluster, the lines above can become a one-liner:

nextflow run nf-core/pangenome -r dev --input references.fa --wfmash_only --wfmash_chunks 5

This emits a results/wfmash folder which stores all the wfmash output.

publications

About

base-accurate DNA sequence alignments using WFA and mashmap2

Resources

Readme
MIT license
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages