Shoujia Zhang

Cancer remains a significant cause of mortality, and the application of microarray technology has opened new avenues for cancer diagnosis and treatment. However, due to the challenges in sample acquisition, the genetic dimension of microarray data surpasses the sample dimension, resulting in high-dimensional small sample data. Effective feature selection is crucial for identifying biomarkers and facilitating further analysis. However, existing methods struggle to fully exploit the interdependencies among genes, such as regulatory networks and pathways, to guide the feature selection process and construct efficient classification models. In this paper, we propose a novel feature selection algorithm and classification model based on graph neural networks to address these challenges. Our proposed method employs a multidimensional graph to capture intricate gene interactions. We leverage link prediction techniques to enhance the graph structure relationships and employ a multidimensional node evaluator alongside a supernode discovery algorithm based on spectral clustering for initial node filtering. Subsequently, a hierarchical graph pooling technique based on downsampling is used to further refine node selection for feature extraction and model building. We evaluate the proposed method on nine publicly available microarray datasets, and the results demonstrate its superiority over both classical and advanced feature selection techniques in various evaluation metrics. This highlights the effectiveness and advancement of our proposed approach in addressing the complexities associated with microarray data analysis and cancer classification.

Background: The discovery of critical biomarkers is significant for clinical diagnosis, drug research and development. Researchers usually obtain biomarkers from microarray data, which comes from the dimensional curse. Feature selection in machine learning is usually used to solve this problem. However, most methods do not fully consider feature dependence, especially the real pathway relationship of genes. Results: Experimental results show that the proposed method is superior to classical algorithms and advanced methods in feature number and accuracy, and the selected features have more significance. Method: This paper proposes a feature selection method based on a graph neural network. The proposed method uses the actual dependencies between features and the Pearson correlation coefficient to construct graph-structured data. The information dissemination and aggregation operations based on graph neural network are applied to fuse node information on graph structured data. The redundant features are clustered by the spectral clustering method. Then, the feature ranking aggregation model using eight feature evaluation methods acts on each clustering sub-cluster for different feature selection. Conclusion: The proposed method can effectively remove redundant features. The algorithm’s output has high stability and classification accuracy, which can potentially select potential biomarkers.