Machine learning models predict the mTOR signal pathway-related signature in the gastric cancer involving 2063 samples of 7 centers

Abstract

Gastric cancer, as a tumor with poor prognosis, has been widely studied. Distinguishing the types of gastric cancer is helpful. Using the transcriptome data of gastric cancer in our study, relevant proteins of mTOR signaling pathway were screened to identify key genes by four machine learning models, and the models were validated in external datasets. Through correlation analysis, we explored the relationship between five key genes and immune cells and immunotherapy. By inducing cellular senescence in gastric cancer cells with bleomycin, we investigated changes in the expression levels of HRAS through western blot. By PCA clustering analysis, we used the five key genes for gastric cancer typing and explored differences in drug sensitivity and enrichment pathways between different clustering groups. We found that the SVM machine learning model was superior, and the five genes (PPARA, FNIP1, WNT5A, HRAS, HIF1A) were highly correlated with different immune cells in multiple databases. These five key genes have a significant impact on immunotherapy. Using the five genes for gastric cancer gene typing, four genes were expressed higher in group 1 and were more sensitive to drugs in group 2. These results suggest that subtype-specific markers can improve the treatment and provide precision drugs for gastric cancer patients.