Biomarker discovery is crucial in advancing personalized medicine by preliminary diagnosis, treatment strategies improvement, and outcome prediction of the disease. Identification of biomarkers through traditional methods is challenging because biomarker identification requires analysis through high-dimensional and complex biological data. In the past, machine learning, in general, and Graph Neural Networks (GNNs) in particular, have made recent strides as a potential alternative by using the context provided by the natural structure of biological networks. In particular, such GNNs are capable of learning complex relationships between biological entities like genes, proteins, and metabolites for more accurate biomarker discovery. This study explores the application of GNNs to biomarker discovery via biomarker discovery to find novel biomarkers that would help in classifying a disease and help in finding a treatment. GNNs are applied to a variety of biological datasets of gene expression and protein interaction data to identify key biological biomarkers of cancer and other diseases. The study also shows that GNNs do vastly better than more traditional machine learning models in predicting much more accurately and formulating a prediction model as well as in formulating a prediction model for properties of real epidermal junctions. This study’s findings suggest that GNNs have the potential to enhance biomarker discovery as a means to improve therapeutics, as well as develop precision medicine. GNNs take on the potential to enable new biomedical research and clinical applications by their ability to handle complex, high-dimensional data and discover hidden patterns within the data. This work establishes the basis of utilizing GNNs to aid in clinical diagnosis and personalization for treating cancers.