A classification method of gastric cancer subtype based on residual graph convolution network
Can Liu1,2, Yuchen Duan1, Qingqing Zhou1
1School of Medical Informatics Engineering, Anhui University of Chinese Medicine, Hefei, Anhui, China.
Frontiers in Genetics
|January 23, 2023
Summary
This study introduces RRGCN, a novel gastric cancer classification model using multi-omics data and patient similarity networks. RRGCN achieves high accuracy, outperforming existing methods for improved cancer diagnosis and treatment insights.
Area of Science:
- Oncology
- Bioinformatics
- Computational Biology
Background:
- Tumor heterogeneity complicates clinical diagnosis and treatment.
- Current cancer subtype classification often overlooks multi-omics data and patient similarities.
Purpose of the Study:
- To develop an advanced gastric cancer subtype classification model.
- To leverage multi-omics fusion data and patient similarity networks for improved accuracy.
Main Methods:
- Developed RRGCN, a residual graph convolutional network (GCN) model.
- Utilized an Autoencoder (AE) for multi-omics data dimensionality reduction and feature extraction.
- Constructed a patient similarity network using Pearson correlation for graph structure.
Main Results:
- RRGCN achieved a classification accuracy of 0.87.
- The model significantly outperformed four traditional machine learning and deep learning methods.
- Demonstrated superior performance in all subtype classification aspects.
Conclusions:
- RRGCN offers a powerful approach for gastric cancer subtype classification.
- The model provides potential new insights into disease mechanisms and progression.
- RRGCN shows promise for broader applications in disease classification and clinical diagnosis.
