PNAbind

PNAbind is a python package and collection of scripts built for computing protein surface meshes, assigning chemical, electrostatic, geometric and MSA features to those meshes, and building/training graph neural network models of protein-nucleic acid binding.

Overview

The overall workflow of our method is shown in the picture above. The fundamental data structure used for representation of a protein is a closed molecular surface mesh. Supervised training data can be generated from protein-nucleic acid complexes for residue-level prediction, or from binding function annotation for protein-level prediction.

Example Features

We provide a large collection of functions for generating electrostatic, chemical, geometrical, and evolutionary features, and functions for mapping arbitrary features defined for atoms, residues, or over a 3D grid to a given surface mesh.

Installation

To run our code, it is recommended to create a new anaconda enviroment and install this package and required dependencies to the fresh environment.

conda create -n pnabind python=3.9
conda activate pndabind

The main dependencies for training and running models are pytorch and torch-geometric. You may try installing the latest versions, but the following should work:

# Choose one of the following that is most appropriate for your system

# CUDA 10.2
conda install pytorch==1.12.1 cudatoolkit=10.2 -c pytorch
# CUDA 11.3
conda install pytorch==1.12.1 cudatoolkit=11.3 -c pytorch
# CUDA 11.6
conda install pytorch==1.12.1 cudatoolkit=11.6 -c pytorch -c conda-forge
# CPU Only (not recommended)
conda install pytorch==1.12.1 cpuonly -c pytorch

conda install pyg==2.1.0 -c pyg

For easy install of additional dependencies, run

pip install -r requirements.txt

Or you can install these dependencies individually as needed (when pnabind throws an exception). Finally, to install the code needed to run our scripts, clone this repository to a local directory and install:

gh repo clone jaredsagendorf/pnabind
cd pnabind
pip install -e .

Soft Requirements

Required for using our mesh and structure feature generation pipeline. The major dependencies are

• Biopython (1.76)+
• trimesh
• NanoShaper or MSMS or EDTSurf available on system path (provided in share)

For electrostatic feature calculations

• APBS or TABI-PB available on system path
• PDB2PQR available on system path

For MSA feature calculations

• PSI-BLAST available on system path
• HHBlits available on system path