A Graph Neural Network Approach

Niccolò Pancino,Caterina Gallegati,Fiamma Romagnoli,Pietro Bongini,Monica Bianchini

DOI: https://doi.org/10.3390/ijms25115870

IF: 5.6

2024-05-29

International Journal of Molecular Sciences

Abstract:Protein–protein interactions (PPIs) are fundamental processes governing cellular functions, crucial for understanding biological systems at the molecular level. Compared to experimental methods for PPI prediction and site identification, computational deep learning approaches represent an affordable and efficient solution to tackle these problems. Since protein structure can be summarized as a graph, graph neural networks (GNNs) represent the ideal deep learning architecture for the task. In this work, PPI prediction is modeled as a node-focused binary classification task using a GNN to determine whether a generic residue is part of the interface. Biological data were obtained from the Protein Data Bank in Europe (PDBe), leveraging the Protein Interfaces, Surfaces, and Assemblies (PISA) service. To gain a deeper understanding of how proteins interact, the data obtained from PISA were assembled into three datasets: Whole, Interface, and Chain, consisting of data on the whole protein, couples of interacting chains, and single chains, respectively. These three datasets correspond to three different nuances of the problem: identifying interfaces between protein complexes, between chains of the same protein, and interface regions in general. The results indicate that GNNs are capable of solving each of the three tasks with very good performance levels.

biochemistry & molecular biology,chemistry, multidisciplinary

What problem does this paper attempt to address?

The paper aims to address key issues in Protein-Protein Interactions (PPIs), particularly by predicting interaction sites between proteins using Graph Neural Networks (GNNs). Specifically, the objectives of the study can be summarized as follows: 1. **PPI Prediction and Site Identification**: Interactions between proteins are crucial for understanding cellular functions. Although experimental methods are reliable, they are limited by time and cost. Therefore, this study explores the use of deep learning methods, especially GNNs, for efficient and cost-effective PPI prediction and site identification. 2. **Application of Graph Neural Networks**: Since protein structures can naturally be represented as graph structures, GNNs are an ideal choice for this task. GNNs can effectively handle graph-structured data, performing feature extraction and classification while preserving spatial information. 3. **Different Levels of PPI Analysis**: To gain a deeper understanding of how proteins interact, the study constructed three different datasets: Whole, Interface, and Chain. These datasets correspond to different levels of the problem: identifying interfaces between protein complexes, interfaces between chains within the same protein, and general interface regions, respectively. 4. **Experimental Results**: The results show that GNNs achieved good performance on all three datasets. They performed best on the Whole dataset, accurately identifying about 81% of interaction nodes while maintaining high precision values. They also achieved good results on the Interface dataset, and although performance declined on the most general Chain dataset, it still reached a good level. In summary, this study is primarily focused on developing a GNN-based method to effectively predict interaction sites between proteins and validating the method's effectiveness and generalization ability through three different levels of datasets.

A Graph Neural Network Approach

Prediction of protein–protein interaction using graph neural networks

Graph Neural Networks for Protein-Protein Interactions -- A Short Survey

Deep Learning Frameworks for Protein–protein Interaction Prediction

Integrating Sequence And Network Information To Enhance Protein-Protein Interaction Prediction Using Graph Convolutional Networks

GNNGL-PPI: multi-category prediction of protein-protein interactions using graph neural networks based on global graphs and local subgraphs

DSSGNN-PPI: A Protein-Protein Interactions prediction model based on Double Structure and Sequence graph neural networks

GraphPI: Efficient Protein Inference with Graph Neural Networks

NPI-GNN: Predicting Ncrna-Protein Interactions with Deep Graph Neural Networks.

Neural Embeddings for Protein Graphs

Structure-aware protein-protein interaction site prediction using deep graph convolutional network

Learning Unknown from Correlations: Graph Neural Network for Inter-novel-protein Interaction Prediction

Decoding the protein-ligand interactions using parallel graph neural networks

An End-to-end Knowledge Graph Fused Graph Neural Network for Accurate Protein-Protein Interactions Prediction

SGPPI: structure-aware prediction of protein–protein interactions in rigorous conditions with graph convolutional network

G- : Knowledge graph neural network for structure-free protein-ligand bioactivity prediction

GGN-GO: geometric graph networks for predicting protein function by multi-scale structure features

Effectiveness and efficiency: label-aware hierarchical subgraph learning for protein-protein interaction

Biological network analysis with deep learning

Predicting Ncrna-Protein Interactions Based on Dual Graph Convolutional Network and Pairwise Learning.

PersGNN: Applying Topological Data Analysis and Geometric Deep Learning to Structure-Based Protein Function Prediction