ANTIPASTI: interpretable prediction of antibody binding affinity exploiting Normal Modes and Deep Learning

ANTIPASTI (ANTIbody Predictor of Affinity from STructural Information) is a Python Deep Learning method that predicts the binding affinity of antibodies from their three-dimensional structure.

Installation

Through PyPI

ANTIPASTI releases are distributed through the Python Package Index (PyPI). To install the latest version use pip:

pip install antipasti

Through Anaconda

We provide an Anaconda environment that satisfies all the dependencies in antipasti-env.yml.

git clone https://github.com/kevinmicha/ANTIPASTI
cd ANTIPASTI
conda env create -f antipasti-env.yml
conda activate antipasti-env
pip install .

Next, you can run the tests to make sure your installation is working correctly.

# While still in the ANTIPASTI directory:
pytest . 

Manually handling the dependencies

If you want to use an existing environment, just omit the Anaconda commands above:

git clone https://github.com/kevinmicha/ANTIPASTI
cd ANTIPASTI
pip install .

or if you need to install it for your user only:

python setup.py install --user 

Requirements

ANTIPASTI requires the following Python packages:

• adabelief-pytorch
• biopython
• matplotlib
• numpy
• opencv-python
• optuna
• pandas
• scikit-learn
• torch
• torchmetrics
• umap-learn

Example Notebooks and Documentation

The full documentation can be found here.

Example notebooks are located in the notebooks folder:

• [Tutorial] Training ANTIPASTI
• [Tutorial] Predicting affinity using ANTIPASTI
• [Tutorial] Explaining binding affinity with ANTIPASTI
• [Tutorial] Combining AlphaFold and ANTIPASTI

You can download normal mode correlation maps computed at atomistic scale (instead of residue scale) here and place them in data/dccm_maps_full_ags_all_aa.

Attribution

If you use this code, please cite the paper indicated in the documentation.