2APGCNN: Advanced Protein Aggregation Prediction Graph Convolutional Neural Network (2APGCNN)

Overview

This repository provides the implementation of a Graph Convolutional Network (GCN) designed to predict protein aggregation (PA) scores. Protein aggregation is a key factor in neurodegenerative diseases such as Alzheimer's and Parkinson's. Our model leverages an enriched dataset sourced from the Protein Data Bank (PDB) and AlphaFold2.0 to achieve high predictive accuracy.

Requirement

We recommend using Python 3.7 or higher. Install the required libraries using the commands below:

pip install torch==1.11.0+cu111 torchvision==0.12.0+cu111 torchaudio==0.8.0 -f https://download.pytorch.org/whl/torch_stable.html
pip install torch-scatter -f https://data.pyg.org/whl/torch-1.11.0+cu102.html
pip install torch-geometric
conda install -y -c rdkit rdkit
pip install biopython
conda install -y -c conda-forge biotite
pip install tqdm

Note: Your environment might differ, so adjust the versions and install any additional libraries as required.

Dataset Preparation

We enhanced the dataset using AGGRESCAN3D 2.0 to calculate PA propensity. Multi-polypeptide chains from PDB data were separated into individual chains, resulting in 302,032 unique entries. Additionally, data from 22,774 Homo sapiens proteins were included from AlphaFold2.0.

Model Performance

Our trained GCN model achieved an outstanding R2 score of 0.99 and a low Mean Absolute Error (MAE), highlighting the effectiveness of incorporating structural data for accurate PA prediction.

Active Learning

An active learning strategy was implemented to identify proteins with high PA propensity efficiently. This approach outperformed other methods, achieving a MAE of 0.0291 in expected improvement and identifying 99% of target proteins by exploring only 29% of the search space.

Citation

If you use this work in your research, please cite our paper [Link to your paper].