Mutual information network-based support vector machine strategy identifies salivary biomarkers in gastric cancer.

PURPOSE: To determine the vital salivary transcriptomic biomarkers for the early detection of gastric cancer via comparing classification efficiency of multiple candidate genes.

METHODS: We firstly identified 5 kinds of candidate genes related to gastric cancer, including differential pathway genes (DPGs) based on the attract method, hub genes in differential pathways based on mutual information network (MIN) analysis, differentially expressed genes (DEGs) identified by Significance Analysis of Microarrays (SAM), informative genes (DEGs in differential pathways), and key genes (hub DEGs). Then, the classification efficiency of these 5 kinds of candidate genes were assessed using support vector machines (SVM) model. The genes with the best classification efficiency were considered as salivary biomarkers in gastric cancer.

RESULTS: Using the attract method, we screened 5 differential pathways in gastric cancer, in which there were 349 DPGs. Based on these DPGs, MIN with 345 genes and 1313 interactions was constructed, from which we obtained 26 hub genes by topological analysis. Meanwhile, we identified 374 DEGs in gastric cancer. Combining DEGs with DPGs and hub genes respectively, we selected 16 informative genes and 5 key genes. SVM analysis showed that the key genes presented the best classification efficiency with AUC=0.99, specificity=1.00, sensitivity=0.98 and MCC=0.95, which would be considered as salivary biomarkers in gastric cancer.

CONCLUSIONS: This study successfully explored several salivary biomarkers for the non-invasive detection of gastric cancer with high specificity and sensitivity, which might contribute to the early detection and treatment of gastric cancer.