AFminD: Predicting binding sites within a big protein

Overview

AFminD is a tool that uses AlphaFold-Multimer to predict binding sites within a big protein. AFminD exploits the fact that the distogram data produced by the AlphaFold-Multimer's distogram head encodes potential interactions between residues. By calculating the residue distances between two protein chains, AFminD can pinpoint potential binding sites. The method is described in the following paper:

Omidi et al., AlphaFold-Multimer accurately captures interactions and dynamics of intrinsically disordered protein regions, Proc. Natl. Acad. Sci. U.S.A., 2024

Installation

You need poetry to be already installed. One way to install it is:

curl -sSL https://install.python-poetry.org | python3 -

Then clone the repository and install the dependencies:

git clone https://github.com/alirezaomidi/AFminD.git
cd AFminD/
poetry install

How to use

Run ColabFold

You need to provide both --zip and --save-all options to ColabFold. The --save-all option will save Disogram head's outputs in .pkl files, which we use later to calculate minD scores. If you use ColabFold's google colab notebook, simply make sure you have chekced the "save_all" option under "Advanced settings".

Calculate expected distances from distogram data

The .pkl files contain full Distogram data in tensors of shape N x N x 64. We need to calculate the expected values for each of the N x N probability distributions. To do so, we use AFminD.compute_expected_distances script:

poetry run python -m AFminD.compute_expected_distances -i /path/to/colabfold/dir/or/zipfile --n-jobs 5

Note that the -i/--input option can take both a directory containing .result.zip files or a single zip file. The script will calculate expected distances and save them in .distogram.json files inside the zip file(s).

Calculate minD values

Expected distances between each pair of residues are now ready. We can calculate the minimum distance between each residue of one chain and any residue from other chains (a.k.a minD):

poetry run python -m AFminD.extract_distogram_scores -i /path/to/colabfold/dir/or/zipfile -o minD.csv --n-jobs 10

Predict binding sites

A residues with low minD is part of a potential binding site. To find them, we should normalize minD values and project them inversely to the range $[0, 1]$. Where minD scores peak, a potential binding site is predicted. Use the following script to do the normalization and peak finding:

poetry run python -m AFminD.find_binding_sites -i minD.csv -o minD_peaks.csv --prominence 0.02 --distance 30

The --prominence and --distance values can be adjusted to your needs. We use $30$ for the distance option to make sure no two peaks are closer than 30 residues, and $0.02$ for the prominence option to make sure no non-significant peaks are found due to noisy signal.

Fragment proteins

Finally, we can cut the protein of interest to the predicted binding site fragments:

poetry run python -m AFminD.cut_binding_sites -f minD_peaks.csv --input /path/to/fasta/file/used/for/colabfold.fasta -o fragments.fasta --window 30 --chain A

The --window option specifies fragment sizes. The --chain option specifies the protein chains to be fragmented. You can specify multiple chains, e.g. --chain A --chain B.

From now, you can use the fragments.fasta file to run ColabFold again and search for possible boosts in ipTM or other metrics shown to be effective in finding potential protein-protein interactions (refer to the reference).

Cite

@article{
    doi:10.1073/pnas.2406407121,
    author = {Alireza Omidi  and Mads Harder Møller  and Nawar Malhis  and Jennifer M. Bui  and Jörg Gsponer },
    title = {AlphaFold-Multimer accurately captures interactions and dynamics of intrinsically disordered protein regions},
    journal = {Proceedings of the National Academy of Sciences},
    volume = {121},
    number = {44},
    pages = {e2406407121},
    year = {2024},
    doi = {10.1073/pnas.2406407121},
    URL = {https://www.pnas.org/doi/abs/10.1073/pnas.2406407121},
    eprint = {https://www.pnas.org/doi/pdf/10.1073/pnas.2406407121},
}