Zhibin Jin ¹ Yuhu Shi ^1* Lili Zhou ²

• ¹ Shanghai Maritime University, pudong, China
• ² Yangpu District Central Hospital, Shanghai, China

While finding the key biomarkers and improving the accuracy of the model, it is equally important to understand the interaction relationship in diseases. In this study, a transparent sparse graph pathway network (TSGPN) is proposed based on the structure of graph neural networks. This network simulates the action of genes in vivo, adds prior knowledge, and improves the accuracy of the model. Firstly, the graph connection is constructed according to protein-protein interaction networks and competing endogenous RNAs (ceRNA) network, and then some noises or unimportant connections are removed spontaneously based on graph attention mechanism and hard concrete estimation, so as to realize the reconstruction of ceRNA network which represents the influence of other genes in the disease on mRNA. Next, the gene-based interpretation is transformed into the pathway-based interpretation based on the pathway database, and the hidden layer is added to realize the high-dimensional analysis of the pathway. Finally, the experimental results show that the proposed TSGPN method is superior to other comparison methods in F1 score and AUC, and more importantly, it can well display the role of genes. Through the data analysis applied to lung cancer prognosis, 10 pathways related to LUSC prognosis were found, as well as the key biomarkers closely related to these pathways, such as HOXA10, hsa-mir-182 and LINC02544. The relationship between them has also been reconstructed to better explain the internal mechanism of disease.