Identifying Cancer Subtypes Using a Residual Graph Convolution Model on a Sample Similarity Network

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Cancer subtype classification is crucial for understanding disease mechanisms and developing targeted therapies.
• Previous methods often overlook the interconnectedness of cancer samples, limiting subtype identification.
• Inter-sample relationships, particularly gene co-expression patterns, can provide valuable insights into cancer heterogeneity.

Purpose of the Study:

• To develop an advanced cancer subtype classification method that leverages sample interactions.
• To improve the accuracy and clinical relevance of cancer subtyping.
• To identify essential genes and biological pathways associated with identified cancer subtypes.

Main Methods:

• A sample similarity network was constructed based on cancer gene co-expression patterns.
• A two-layer graph convolutional network (GCN) model integrated gene expression profiles and the sample similarity network.
• Initial features were incorporated into the GCN to mitigate over-smoothing, followed by softmax classification.

Main Results:

• The proposed model achieved high accuracy in classifying subtypes for breast invasive carcinoma (BRCA, 82.58%), glioblastoma multiforme (GBM, 85.13%), and lung cancer (LUNG, 79.18%).
• Performance surpassed existing cancer subtype classification methods.
• Survival analysis confirmed the clinical significance of the identified subtypes.

Conclusions:

• The developed GCN-based method effectively classifies cancer subtypes by considering sample similarities and gene expression.
• The model's ability to identify essential genes and pathways offers potential for novel therapeutic strategies.
• This approach enhances our understanding of cancer pathogenesis and facilitates personalized treatment strategies.