MMseqs2 (Many-against-Many sequence searching) is a software suite to search and cluster huge protein and nucleotide sequence sets. MMseqs2 is free and open source software implemented in C++ for Linux, MacOS, and (as beta version, via cygwin) Windows. The software is designed to run on multiple cores and servers and exhibits very good scalability. MMseqs2 can run 10000 times faster than BLAST. At 100 times its speed it achieves almost the same sensitivity. It can perform profile searches with the same sensitivity as PSI-BLAST at over 400 times its speed.
The MMseqs2 user guide is available in our GitHub Wiki or as a PDF file (Thanks to pandoc!). The wiki also contains tutorials to learn how to use MMseqs2 with real data. For questions please open an issue on GitHub. Keep posted about MMseqs2/Linclust updates by following Martin on Twitter.
MMseqs2 can be used by compiling from source, downloading a statically compiled binary at mmseqs.com/latest, using Homebrew, conda or Docker.
# install by brew
brew install mmseqs2
# install via conda
conda install -c conda-forge -c bioconda mmseqs2
# install docker
docker pull ghcr.io/soedinglab/mmseqs2
# MMseqs2-GPU mostly-static AVX2 build requiring glibc >= 2.29 and nvidia driver >=525.60.13 (see below)
wget https://mmseqs.com/latest/mmseqs-linux-gpu.tar.gz; tar xvfz mmseqs-linux-gpu.tar.gz; export PATH=$(pwd)/mmseqs/bin/:$PATH
# static build with AVX2 (fastest)
wget https://mmseqs.com/latest/mmseqs-linux-avx2.tar.gz; tar xvfz mmseqs-linux-avx2.tar.gz; export PATH=$(pwd)/mmseqs/bin/:$PATH
# static build with SSE4.1
wget https://mmseqs.com/latest/mmseqs-linux-sse41.tar.gz; tar xvfz mmseqs-linux-sse41.tar.gz; export PATH=$(pwd)/mmseqs/bin/:$PATH
# static build with SSE2 (slowest, for very old systems)
wget https://mmseqs.com/latest/mmseqs-linux-sse2.tar.gz; tar xvfz mmseqs-linux-sse2.tar.gz; export PATH=$(pwd)/mmseqs/bin/:$PATH
MMseqs2 requires an AMD or Intel 64-bit system (check with uname -a | grep x86_64). We recommend using a system with at least the SSE4.1 instruction set (check by executing cat /proc/cpuinfo | grep sse4_1 on Linux or sysctl -a | grep machdep.cpu.features | grep SSE4.1 on MacOS). The AVX2 version is faster than SSE4.1, check if AVX2 is supported by executing cat /proc/cpuinfo | grep avx2 on Linux and sysctl -a | grep machdep.cpu.leaf7_features | grep AVX2 on MacOS. A SSE2 version is also available for very old systems. MMseqs2 also works on ARM64 systems and on PPC64LE systems with POWER8 ISA or newer.
Note
We recently added support for GPU-accelerated protein sequence and profile searches. This requires an NVIDIA GPU of the Ampere generation or newer for full speed, however, also works at reduced speed for Tesla-generation GPUs. Check the wiki for instructions on how to get started.
MMseqs2 comes with a bash command and parameter auto completion, which can be activated by adding the following to your $HOME/.bash_profile:
if [ -f /Path to MMseqs2/util/bash-completion.sh ]; then
source /Path to MMseqs2/util/bash-completion.sh
fi
We provide easy workflows to cluster, search and assign taxonomy. These easy workflows are a shorthand to deal directly with FASTA/FASTQ files as input and output. MMseqs2 provides many modules to transform, filter, execute external programs and search. However, these modules use the MMseqs2 database formats, instead of the FASTA/FASTQ format. For maximum flexibility, we recommend using MMseqs2 workflows and modules directly. Please read more about this in the documentation.
For clustering, MMseqs2 easy-cluster and easy-linclust are available.
easy-cluster by default clusters the entries of a FASTA/FASTQ file using a cascaded clustering algorithm.
mmseqs easy-cluster examples/DB.fasta clusterRes tmp --min-seq-id 0.5 -c 0.8 --cov-mode 1
easy-linclust clusters the entries of a FASTA/FASTQ file. The runtime scales linearly with input size. This mode is recommended for huge datasets.
mmseqs easy-linclust examples/DB.fasta clusterRes tmp
Read more about the clustering format in our user guide.
Please adjust the clustering criteria and check if temporary directory provides enough free space. For disk space requirements, see the user guide.
The easy-search workflow searches directly with a FASTA/FASTQ files against either another FASTA/FASTQ file or an already existing MMseqs2 database.
mmseqs easy-search examples/QUERY.fasta examples/DB.fasta alnRes.m8 tmp
It is also possible to pre-compute the index for the target database. This reduces overhead when searching repeatedly against the same database.
mmseqs createdb examples/DB.fasta targetDB
mmseqs createindex targetDB tmp
mmseqs easy-search examples/QUERY.fasta targetDB alnRes.m8 tmp
The databases workflow provides download and setup procedures for many public reference databases, such as the Uniref, NR, NT, PFAM and many more (see Downloading databases). For example, to download and search against a database containing the Swiss-Prot reference proteins run:
mmseqs databases UniProtKB/Swiss-Prot swissprot tmp
mmseqs easy-search examples/QUERY.fasta swissprot alnRes.m8 tmp
The speed and sensitivity of the search can be adjusted with -s parameter and should be adapted based on your use case (see setting sensitivity -s parameter). A very fast search would use a sensitivity of -s 1.0, while a very sensitive search would use a sensitivity of up to -s 7.0. A detailed guide how to speed up searches is here.
The output can be customized with the --format-output option e.g. --format-output "query,target,qaln,taln" returns the query and target accession and the pairwise alignments in tab separated format. You can choose many different output columns.
❗ easy-search in default computes the sequence identity by dividing the number of identical residues by the alignment length (numIdentical/alnLen). However, search estimates the identity in default. To output real sequence identity use --alignment-mode 3 or -a.
The easy-taxonomy workflow can be used to assign sequences taxonomical labels. It performs a search against a sequence database with taxonomy information (seqTaxDb), chooses the most representative sets of aligned target sequences according to different strategies (according to --lca-mode) and computes the lowest common ancestor among those.
mmseqs createdb examples/DB.fasta targetDB
mmseqs createtaxdb targetDB tmp
mmseqs createindex targetDB tmp
mmseqs easy-taxonomy examples/QUERY.fasta targetDB alnRes tmp
By default, createtaxdb assigns a Uniprot accession to a taxonomical identifier to every sequence and downloads the NCBI taxonomy. We also support BLAST, SILVA or custom taxonomical databases. Many common taxonomic reference databases can be easily downloaded and set up by the databases workflow.
Read more about the taxonomy format and the classification in our user guide.
MMseqs2 provides many additional search modes:
- Iterative sequences-profile searches (like PSI-BLAST) with the
--num-iterationsparameter - Translated searches of nucleotides against proteins (blastx), proteins against nucleotides (tblastn) or nucleotide against nucleotide (tblastx)
- Iterative increasing sensitivity searches to find only the best hits faster
- Taxonomic assignment using 2bLCA or LCA
- Fast ungapped alignment searches to find very similar sequence matches
- Very fast and sensitive searches against profile databases such as the PFAM
- Reciprocal best hits search
- Web search API and user interface
Many modes can also be combined. You can, for example, do a translated nucleotide against protein profile search.
MMseqs2 minimum memory requirements for cluster or linclust is 1 byte per sequence residue, search needs 1 byte per target residue. Sequence databases can be compressed using the --compress flag, DNA sequences can be reduced by a factor of ~3.5 and proteins by ~1.7.
MMseqs2 checks the available system memory and automatically divides the target database in parts that fit into memory. Splitting the database will increase the runtime slightly. It is possible to control the memory usage using --split-memory-limit.
MMseqs2 can run on multiple cores and servers using OpenMP and Message Passing Interface (MPI). MPI assigns database splits to each compute node, which are then computed with multiple cores (OpenMP).
Make sure that MMseqs2 was compiled with MPI by using the -DHAVE_MPI=1 flag (cmake -DHAVE_MPI=1 -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=. ..). Our precompiled static version of MMseqs2 cannot use MPI. The version string of MMseqs2 will have a -MPI suffix, if it was built successfully with MPI support.
To search with multiple servers, call the search or cluster workflow with the MPI command exported in the RUNNER environment variable. The databases and temporary folder have to be shared between all nodes (e.g. through NFS):
RUNNER="mpirun -pernode -np 42" mmseqs search queryDB targetDB resultDB tmp
MMseqs2 exists thanks to all the people who contribute.