Risk stratification and pathway analysis based on graph neural network and interpretable algorithm.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Pathway-based analysis of transcriptomic data offers advantages over gene-based methods.
• Existing deep learning models for pathway analysis often neglect crucial pathway topological features, limiting predictive accuracy.

Purpose of the Study:

• To introduce PathGNN, a novel Graph Neural Network (GNN) model designed to integrate pathway topology for enhanced bioinformatic analysis.
• To evaluate PathGNN's performance in predicting long-term cancer survival.

Main Methods:

• Construction of a Graph Neural Network (GNN) model (PathGNN) to capture pathway topological features.
• Application of PathGNN to predict long-term survival in four distinct cancer types.
• Utilizing an interpretation algorithm to identify survival-associated pathways.

Main Results:

• PathGNN demonstrated promising predictive performance for long-term cancer survival.
• The model's performance surpassed that of other commonly used methods.
• The interpretation algorithm successfully identified relevant pathways linked to survival outcomes.

Conclusions:

• Graph Neural Networks (GNNs) are effective for developing pathway-based models in bioinformatics.
• PathGNN offers a powerful approach for improving predictive accuracy in cancer survival analysis.
• The model's ability to identify key pathways enhances biological interpretability.