Cancer subtype classification and modeling by pathway attention and propagation

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Biological pathways are crucial for understanding cellular processes and cancer mechanisms.
• Current computational methods struggle to effectively utilize fragmented pathway data for cancer subtype classification.

Purpose of the Study:

• To develop an explainable deep-learning model for cancer subtype classification using biological pathways.
• To address the limitations of existing methods in leveraging pathway information.

Main Methods:

• Utilized graph convolutional networks to model individual pathways.
• Employed a multi-attention ensemble model to integrate hundreds of pathways.
• Incorporated network propagation to explain subtype-specific gene expression differences.

Main Results:

• Achieved high classification accuracies across five TCGA cancer datasets.
• Successfully identified subtype-specific pathways and their associated biological functions.
• The model provides explainable insights into cancer subtype distinctions.

Conclusions:

• The developed deep-learning model offers a robust and explainable approach for cancer subtype classification.
• This method enhances the understanding of biological mechanisms underlying different cancer subtypes.
• The approach effectively leverages pathway information, overcoming previous computational limitations.