ESP_DNN Graph Convolutional Deep Neural Network for Electrostatic Potential Surface Prediction in DFT(MATLAB Source Code)

ESP-DNN: Graph Convolutional Deep Neural Network for Predicting Electrostatic Potential Surfaces from DFT Calculations

This repository contains trained models and code designed for generating ligands and proteins, creating electrostatic potential (ESP) surfaces that closely resemble DFT-quality molecular surfaces. The PQR files generated by our model include atomic charges and dipole-like atomic features, such as lone pairs, σ-conjugation, and p-orbitals. To generate ligand PQR files, a graph convolutional deep neural network (DNN) model was trained on about 100,000 molecules with ESP surfaces derived from DFT calculations.

For proteins, parameterized charges of amino acids were used, ensuring compatibility with the ligand ESP surfaces generated by the DNN model. For more detailed methods and validation information, refer to the full documentation.

System Requirements

The program can only run on 64-bit Linux operating systems.

Installation Instructions

To run ESP-DNN, you will need to:

1. Clone this repository.

2. Set up Python and required dependencies.

3. (Optional) Install additional packages.

The package has been developed and tested with Python 2.7 and the following third-party libraries:

- rdkit == 2018.09.3

- keras == 2.2.4

- tensorflow == 1.10.0

- num