Use of machine learning-based integration to develop an immune-related signature for improving prognosis in patients with gastric cancer

Abstract

Abstract Gastric cancer is one of the most common malignancies. Although some patients benefit from immunotherapy, the majority of patients have unsatisfactory immunotherapy outcomes, and the clinical significance of immune-related genes in gastric cancer remains unknown. We used the single-sample gene set enrichment analysis (ssGSEA) method to evaluate the immune cell content of gastric cancer patients from TCGA and clustered patients based on immune cell scores. The Weighted Correlation Network Analysis (WGCNA) algorithm was used to identify immune subtype-related genes. The patients in TCGA were randomly divided into test 1 and test 2 in a 1:1 ratio, and a machine learning integration process was used to determine the best prognostic signatures in the total cohort. The signatures were then validated in the test 1 and the test 2 cohort. Based on a literature search, we selected 93 previously published prognostic signatures for gastric cancer and compared them with our prognostic signatures. At the single-cell level, the algorithms "Seurat," "SCEVAN", "scissor", and "Cellchat" were used to demonstrate the cell communication disturbance of high-risk cells. WGCNA and univariate Cox regression analysis identified 52 prognosis-related genes, which were subjected to 98 machine-learning integration processes. A prognostic signature consisting of 24 genes was identified using the StepCox[backward] and Enet[alpha = 0.7] machine learning algorithms. This signature demonstrated the best prognostic performance in the overall, test1 and test2 cohort, and outperformed 93 previously published prognostic signatures. Interaction perturbations in cellular communication of high-risk T cells were identified at the single-cell level, which may promote disease progression in patients with gastric cancer. We developed an immune-related prognostic signature with reliable validity and high accuracy for clinical use for predicting the prognosis of patients with gastric cancer.