Augmented Potts

The Augmented Potts supervised regression model for variant effect prediction as first described in: Learning protein fitness models from evolutionary and assay-labeled data and exactly matching the implementation of the respective model used in: Enzyme Structure Correlates With Variant Effect Predictability

Installation

augmented-potts calls GMRELIN_CPP to fit the Markov random field to the MSA and generate emission parameters used to compute sequence energies. Make sure that GREMLIN_CPP is installed and added to $PATH.

Using pip

git clone https://github.com/florisvdf/augmented-potts.git
cd augmented-potts
pip install .

Using uv

git clone https://github.com/florisvdf/augmented-potts.git
cd augmented-potts
uv sync

Usage

Train the Augmented Potts model with the entrypoint script:

$ python src/augmented_potts/run.py --help
Usage: run.py [OPTIONS] DATA_PATH TARGET

╭─ Arguments ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ *    data_path      TEXT  [default: None] [required]                                                                                             │
│ *    target         TEXT  [default: None] [required]                                                                                             │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭─ Options ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ --potts-path        TEXT   [default: None]                                                                                                       │
│ --msa-path          TEXT   [default: None]                                                                                                       │
│ --encoder           TEXT   [default: energies]                                                                                                   │
│ --alpha             FLOAT  [default: 0.1]                                                                                                        │
│ --output-dir        TEXT   [default: None]                                                                                                       │
│ --help                     Show this message and exit.                                                                                           │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

Or import AugmentedPotts and train the model in your own script:

from augmented_potts.model import AugmentedPotts


model = AugmentedPotts(msa_path=my_msa_path)
model.fit(data=my_dataframe, target=my_target)

The fit method of the AugmentedPotts expects a dataframe with a sequence column storing protein variant sequences of identical length and will train on values in the column matching the string passed to the target argument.