This is a new Dockerfile that allows the use of MGKit using a container. You can run the scripts directly, for example:
docker run --rm -i frubino/mgkit:latest sampling-utils rand_seq
Will run the sampling-utils rand_seq to create some randome FASTA sequences. Commands can be piped as well:
docker run --rm -i frubino/mgkit:latest sampling-utils rand_seq | docker run --rm -i frubino/mgkit:latest fasta-utils translate
Will translate the random sequneces from the first command. Highly suggested to use an alias, such as:
alias mgkit='docker run --rm -i frubino/mgkit:latest'
This way the above command becomes:
mgkit sampling-utils rand_seq | mgkit fasta-utils translate
If you want to run interactively a series of commands you can use bash instead of another command, but remember to add the -t option:
docker run --rm -it frubino/mgkit:latest bash
A frubino/mgkit:build branch and image is present to allow the creation of Conda packages. Checkout the branch with git checkout build. A script is included to build the image and environment are used to specify output directory inside the container, the Python version to use to build and the MGKit version to use
Content type
Image
Digest
Size
938.4 MB
Last updated
almost 5 years ago
docker pull frubino/mgkit