Identifying Cancer Stage-Related Biomarkers for Lung Adenocarcinoma by Integrating Both Node and Edge Features

Background: In order to characterize phenotypes and diseases, genetic factors and their interactions in biological systems must be considered. Although genes or node features are the core units of genetic information, their connections, also known as edge features, are composed of a network of gene interactions. These components are crucial for understanding the molecular basis of disease and phenotype development. Existing research typically utilizes node biomarkers composed of individual genes or proteins for the binary classification of cancer. However, due to significant heterogeneity among patients, these methods cannot adapt to the subtle changes required for precise cancer staging, and relying solely on node biomarkers often leads to poor accuracy in classifying cancer staging. Methods: In this study, a computational framework was developed to diagnose lung adenocarcinoma, integrating node and edge features such as correlation, covariance, and residuals. The proposed method allows for precise diagnosis in the case of a single sample, which can identify the minimum feature set that effectively distinguishes cancer staging. Results: The advantages of the proposed method are: (i) it can diagnose each individual test sample, promoting personalized treatment; (ii) integrating node and edge features can improve diagnostic accuracy, indicating that each type of feature can capture unique aspects of the disease; (iii) it significantly reduces the number of features required to accurately classify the four stages of cancer, thereby achieving optimal cross-validation accuracy. Conclusions: This streamlined and effective feature set highlights the potential of our approach in advancing personalized medicine and improving clinical outcomes for cancer patients.