Prediction of amino acid substitution impact using homologs and a directional substitution matrix.

DeMaSk predictions can be easily obtained for any protein sequence using the web tool. This package can be downloaded for customized usage. See the full documentation for more detailed instructions.

To install, clone the repository or download and unzip. To get Python dependencies and be able to run or import the modules from any directory, install with pip (or pip3 if pip = pip2):

pip install -e DeMaSk/

Unless you're supplying your own aligned homologs, you'll need to have the blastp program. It can be downloaded as part of BLAST+.

The blastp step requires a formatted sequence database. To use UniRef90, which demask.princeton.edu uses, download the zipped fasta file from here, unzip, and use makeblastdb from BLAST+ to format the database.

To avoid having to specify the location of the blastp binary and the database for every DeMaSk run, put them in a config file, e.g.:

blastp=/usr/local/ncbi/blast/bin/blastp
db=/path/to/database/uniref90.fasta

By default, DeMaSk will look for the config file DeMaSk/config.ini, which can also contain any other command line arguments, such as nseqs, threads, and matrix.

Run any of the command modules with -h to see all options, e.g.:

python3 -m demask.homologs -h

Once the demask package is installed, you can run run it from anywhere. If you don't have aligned homologs for your query yet, run the demask.homologs module:

python3 -m demask.homologs -s myquery.fa -o myquery_homologs.a2m

The above command will also produce a file myquery.blast.json containing the intermediate blastp output.

Then, generate fitness impact predictions for all single-residue variants of the query sequence:

python3 -m demask.predict -i myquery_homologs.a2m -o myquery_predictions.txt

The output looks like this:

pos   WT      var     score   entropy log2f_var       matrix
1     M       A       -0.3019 1.0666  -21.9207        -0.2641
1     M       C       -0.3074 1.0666  -21.9207        -0.2713
1     M       D       -0.4134 1.0666  -21.9207        -0.4100
1     M       E       -0.4036 1.0666  -21.9207        -0.3972
1     M       F       -0.2183 1.0666  -4.5455         -0.2707
1     M       G       -0.3828 1.0666  -21.9207        -0.3700
...

Corresponding functions can be run in Python code by importing demask.homologs.find_homologs and demask.predict.run_demask.

DeMaSk comes with a directional substitution matrix computed from a collection of deep mutational scanning datasets, as well as corresponding linear model coefficients. Additional commands are included in case you want to fit the model to a custom matrix, or even calculate a matrix from a custom data collection and then fit the linear model to it.

For example, the default matrix was generated like this:

python3 -m demask.matrix -d DeMaSk/data/datasets -o DeMaSk/data/matrix.txt

Then, linear model coefficients were calculated:

python3 -m demask.fit -d DeMaSk/data/datasets -a DeMaSk/data/alignments -m DeMaSk/data/matrix.txt -o DeMaSk/data/coefficients.txt

Corresponding functions can be run in Python code by importing demask.matrix.prepare_matrix and demask.fit.fit_model.