Warning: This guide is not for setting up ColabFold. Please refer to the instructions in the ColabFold repository for setting up a MSA server.
Install docker, docker-compose and git on your server.
If you want to run multiple worker instances, make sure to either use the AUFS or overlay/overlay2 storage drivers. Without these, the database indices will be held multiple times in main memory and performance will degrade.
To check out the repository, run:
git clone https://github.com/soedinglab/mmseqs2-app
cd mmseqs2-app/docker-compose
If you want to use one of our default databases (Uniclust/Pfam/PDB/EggNOG) go to section Setting Up Custom Databases then return here. If you want to use your own sequence databases go to section Setting up a Custom Sequence Database then return here. If you want to use your own profile databases go to section Setting up a Custom Profile Database then return here.
Take a look at the MMseqs2 User Guide to make sure you have enough system memory for all databases.
Once you have set up your search databases you can start the MMseqs2 Search Server by executing:
docker-compose up
You can now navigate with a web browser to your server's IP address and use the MMseqs2 Search Server.
To make sure you are running the latest version of the MMseqs2 Web Server execute the following commands.
docker-compose pull
docker-compose up
Execute the provided docker-compose script to initiate database setup for the web server. This script leverages the mmseqs or foldseek database workflows:
# To list all available databases:
docker-compose run db-setup
# To set up specific databases, e.g., UniRef50 and PDB70:
docker-compose run db-setup UniRef50 PDB70For reference purposes, previous setup scripts are located in the example folder.
We provide a script in examples/uniclust/setup.sh that will download and setup the Uniclust sequence database for you.
Please take a look at this script and choose version, sequence identity level and kind of the database you want.
We provide databases at 30%, 50% and 90% sequence identity, both as seed databases (the cluster reference sequence is the original Uniprot header and sequence) and as a consensus database (the cluster reference sequence is a Uniclust consensus sequence with a summarized header).
After you choose the parameters to your liking, navigate to the databases folder and execute the setup script:
cd databases
../examples/uniclust/setup.sh
When the script finishes, you can optionally edit the generated params file. Take a look at the "Params File" section.
We provide a script in examples/pdb70/setup.sh that will download and setup the PDB70 profile database for you.
After you choose the parameters to your liking, navigate to the databases folder and execute the setup script:
cd databases
../examples/pdb70/setup.sh
We provide a script in examples/pfam/setup.sh that will download and setup the Pfam profile database for you.
After you choose the parameters to your liking, navigate to the databases folder and execute the setup script:
cd databases
../examples/pfam/setup.sh
We provide a script in examples/eggnog/setup.sh that will download and setup the EggNOG profile database for you.
Since building the EggNOG database is slightly more involved, we use a docker container to set it up.
cd examples/eggnog
./setup.sh
To add a new sequence database to your local MMseqs2 Search Server, place your sequences as a FASTA file with the .fasta file ending in the databases folder.
Then create a .params file with the same basename (filename without the .fasta ending) in the same folder.
The following command will create an empty .params file that you can customize:
BASEFASTA="sequence_db"
echo -e "{\n \"name\": \"\",\n \"version\": \"\",\n \"path\": \"${BASEFASTA}\",\n \"default\": true,\n \"order\": 0,\n \"index\": \"\",\n \"search\": \"\"\n}" > "${BASEFASTA}.params"
Take a look at the "Params File" to customize this file.
To add a profile database to your local MMseqs2 Search Server, you need to provide multiple sequence alignments for each target profile inside an MMseqs2 indexed database. Place your MSAs as one STOCKHOLM file with the .sto file ending in the databases folder.
The params file contains a JSON object file describing each search database with the following properties:
{
"name" : "Uniclust30", // string Human readable name of the database
"version" : "2018_08", // string Description of the database, usually the version number
"path" : "", // string (optional) Database base name in the config databases folder
// will be automatically resolved and thus does not need to be specified
"default" : true, // bool Should the database be selected by default
"order" : 0, // int Order in which the databases appear in the frontend
"taxonomy": false, // bool Does the database contain taxonomic annotations
"index" : "-s 6", // string (optional)
// Additional parameters for MMseqs2's createindex module
// -s FLOAT Decides how many k-mers to store per target in the search database
// Sensitivity in actual search must by <= than this value (default 5.7)
// See mmseqs createindex -h for a list of all parameters
// Multiple parameters should be given as a single string: E.g. "-s 6 -v 0"
"search" : "-s 6" // string
// Additonal parameters for MMseqs2's easy-search module
// Useful parameters:
// --search-type In nucleotide-nucleotide searches:
// 2) Use translated protein to translated protein search
// 3) Use nucleotide to nucleotide search
// --num-iterations INT values > 1 will call the iterative-profile search
// -s FLOAT Search sensitivity decides how many k-mers in the prefiltering.
// Must by <= than the value given to index (default 5.7)
// See mmseqs easy-search -h for a list of all parameters
// Multiple parameters should be given as a single string: E.g. "-s 6 -v 0"
"status" : "PENDING", // PENDING|RUNNING|COMPLETE|ERROR|UNKNOWN (optional)
// Current status of the database, will be PENDING for a new database
// If server crashes while building the database it might be stuck in RUNNING state.
// Manually reset it to PENDING in this case
}
If you want to customize the application, you can get docker to build all images locally, instead of fetching them from the registry, by running:
docker-compose -f docker-compose.yml -f docker-compose-dev.yml up