Important Information

We are happy to announce that the index size on disk is down by 8 times and in memory by 4 times due to moving to only one type of FM-index (2bit.64 instead of 2bit.64 and 8bit.32) and 8x compression of suffix array. For example, for human genome, index size on disk is down to ~10GB from ~80GB and memory footprint is down to ~10GB from ~40GB. There is a substantial reduction in index IO time due to the reduction and hardly any performance impact on read mapping. Due to this change in index structure (in commit #4b59796, 10th October 2020), you will need to rebuild the index.

Added MC flag in the output sam file in commit a591e22. Output should match original bwa-mem version 0.7.17.

As of commit e0ac59e, we have a git submodule safestringlib. To get it, use --recursive while cloning or use "git submodule init" and "git submodule update" in an already cloned repository (See below for more details).

Getting Started

# Use precompiled binaries (recommended)
curl -L https://github.com/bwa-mem2/bwa-mem2/releases/download/v2.2.1/bwa-mem2-2.2.1_x64-linux.tar.bz2 \
  | tar jxf -
bwa-mem2-2.2.1_x64-linux/bwa-mem2 index ref.fa
bwa-mem2-2.2.1_x64-linux/bwa-mem2 mem ref.fa read1.fq read2.fq > out.sam

# Compile from source (not recommended for general users)
# Get the source
git clone --recursive https://github.com/bwa-mem2/bwa-mem2
cd bwa-mem2
# Or
git clone https://github.com/bwa-mem2/bwa-mem2
cd bwa-mem2
git submodule init
git submodule update
# Compile and run
make
./bwa-mem2

Introduction

The tool bwa-mem2 is the next version of the bwa-mem algorithm in bwa. It produces alignment identical to bwa and is ~1.3-3.1x faster depending on the use-case, dataset and the running machine.

The original bwa was developed by Heng Li (@lh3). Performance enhancement in bwa-mem2 was primarily done by Vasimuddin Md (@yuk12) and Sanchit Misra (@sanchit-misra) from Parallel Computing Lab, Intel. bwa-mem2 is distributed under the MIT license.

Installation

For general users, it is recommended to use the precompiled binaries from the release page. These binaries were compiled with the Intel compiler and runs faster than gcc-compiled binaries. The precompiled binaries also indirectly support CPU dispatch. The bwa-mem2 binary can automatically choose the most efficient implementation based on the SIMD instruction set available on the running machine. Precompiled binaries were generated on a CentOS7 machine using the following command line:

make CXX=icpc multi

Usage

The usage is exactly same as the original BWA MEM tool. Here is a brief synopsys. Run ./bwa-mem2 for available commands.

# Indexing the reference sequence (Requires 28N GB memory where N is the size of the reference sequence).
./bwa-mem2 index [-p prefix] <in.fasta>
Where 
<in.fasta> is the path to reference sequence fasta file and 
<prefix> is the prefix of the names of the files that store the resultant index. Default is in.fasta.

# Mapping 
# Run "./bwa-mem2 mem" to get all options
./bwa-mem2 mem -t <num_threads> <prefix> <reads.fq/fa> > out.sam
Where <prefix> is the prefix specified when creating the index or the path to the reference fasta file in case no prefix was provided.

Performance

Datasets:
Reference Genome: human_g1k_v37.fasta

Alias	Dataset source	No. of reads	Read length
D1	Broad Institute	2 x 2.5M bp	151bp
D2	SRA: SRR7733443	2 x 2.5M bp	151bp
D3	SRA: SRR9932168	2 x 2.5M bp	151bp
D4	SRA: SRX6999918	2 x 2.5M bp	151bp

Machine details:
Processor: Intel(R) Xeon(R) 8280 CPU @ 2.70GHz
OS: CentOS Linux release 7.6.1810
Memory: 100GB

We followed the steps below to collect the performance results:
A. Data download steps:

1. Download SRA toolkit from https://trace.ncbi.nlm.nih.gov/Traces/sra/sra.cgi?view=software#header-global
2. tar xfzv sratoolkit.2.10.5-centos_linux64.tar.gz
3. Download D2: sratoolkit.2.10.5-centos_linux64/bin/fastq-dump --split-files SRR7733443
4. Download D3: sratoolkit.2.10.5-centos_linux64/bin/fastq-dump --split-files SRR9932168
5. Download D4: sratoolkit.2.10.5-centos_linux64/bin/fastq-dump --split-files SRX6999918

B. Alignment steps:

1. git clone https://github.com/bwa-mem2/bwa-mem2.git
2. cd bwa-mem2
3. make CXX=icpc (using intel C/C++ compiler)
   or make (using gcc compiler)
4. ./bwa-mem2 index <ref.fa>
5. ./bwa-mem2 mem [-t <#threads>] <ref.fa> <in_1.fastq> [<in_2.fastq>] > <output.sam>

For example, in our double socket (56 threads each) and double numa compute node, we used the following command line to align D2 to human_g1k_v37.fasta reference genome.

numactl -m 0 -C 0-27,56-83 ./bwa-mem2 index human_g1k_v37.fasta  
numactl -m 0 -C 0-27,56-83 ./bwa-mem2 mem -t 56 human_g1k_v37.fasta SRR7733443_1.fastq SRR7733443_2.fastq > d2_align.sam

bwa-mem2 seeding phase accelerated using LISA (Learned-Indexes for Sequence Analysis)

bwa-mem2-lisa is an accelerated version of bwa-mem2 where we apply learned-indexes to the seeding phase. bwa-mem2-lisa branch contains the source code of the implementation. Following are the features of bwa-mem2-lisa:

1. Exact same output as bwa-mem2.
2. All command-lines for creating an index and the read mapping are exactly same as bwa-mem2.
3. bwa-mem2-lisa accelerates seeding phase (one of the major bottlenecks in bwa-mem2) by up to 4.5x compared to bwa-mem2.
4. The memory footprint of bwa-mem2-lisa index is ~120GB for human genome.
5. The code is present in bwa-mem2-lisa branch: https://github.com/bwa-mem2/bwa-mem2/tree/bwa-mem2-lisa

bwa-mem2 seeding speedup with Enumerated Radix Trees (Code in ert branch)

The ert branch of bwa-mem2 repository contains codebase of enuerated radix tree based acceleration of bwa-mem2. The ert code is built on the top of bwa-mem2 (thanks to the hard work by @arun-sub). The following are the highlights of the ert based bwa-mem2 tool:

1. Exact same output as bwa-mem(2)
2. The tool has two additional flags to enable the use of ert solution (for index creation and mapping), else it runs in vanilla bwa-mem2 mode
3. It uses 1 additional flag to create ert index (different from bwa-mem2 index) and 1 additional flag for using that ert index (please see the readme of ert branch)
4. The ert solution is 10% - 30% faster (tested on above machine configuration) in comparison to vanilla bwa-mem2 -- users are adviced to use option -K 1000000 to see the speedups
5. The memory foot print of the ert index is ~60GB
6. The code is present in ert branch: https://github.com/bwa-mem2/bwa-mem2/tree/ert

Citation

Vasimuddin Md, Sanchit Misra, Heng Li, Srinivas Aluru. Efficient Architecture-Aware Acceleration of BWA-MEM for Multicore Systems. IEEE Parallel and Distributed Processing Symposium (IPDPS), 2019. 10.1109/IPDPS.2019.00041