Machine Learning Algorithm-Based Screening of Bladder Tumor Prognosis-Related Genes and Analysis of Their Re-Lationship with Immune Infiltrating Cells

Zhou Yuhe,Wang Chengyu,Fu Bo,Huang zhenglin,Ning Xianghui,Jia Zhankui,Yang Jinjian

DOI: https://doi.org/10.16193/j.cnki.hnwk.2023.01.039

2023-01-01

Abstract:Objective Screening for prognostic-related genes in bladder tumors and analyzing their relationship with immune infiltrating cells.Methods The microarray data were obtained from the Gene Expression Omnibus（GEO） database as the training set was used to identify differentially expressed genes（DEGs）. Signature genes were screened using LASSO regression and SVM-RFE algorithms. Bladder tumor RNA-seq FPKM data downloaded from The Cancer Genome Atlas Program database（TCGA） was used as the validation set to verify the sensitivity and specificity of key genes. TCGA clinical data were downloaded for survival analysis of key genes and validated with the GEPIA database. The CIBERSORT website was used to perform immune infiltrating cell analysis on the validation set data and to explore the relationship between prognosis-related genes and immune infiltrating cells.Results In the training group, 176 DEGs were obtained, including 21 up-regulated genes and 155 down-regulated genes. The LASSO regression and SVM-RFE algorithms obtained 13, 34 feature genes, respectively, and five key genes（CYTL1, SRPX, ADH1B, PMP2, IQGAP3） were obtained after taking the intersection. The key genes all showed high sensitivity and specificity in the validation group. Survival analysis revealed significant differences in the survival rates of PMP2（P=0.002）, CYTL1（P=0.032） genes with different expression levels. In tumor tissues, M0 macrophages and M1 macrophages showed high levels; resting mast cells showed low levels. PMP2 and CYTL1 expressions were negatively correlated with T cell follicular helper, CD8 T cells, and positively correlated with resting memory CD4 T cells, naive B cells. The patients with high abundance of activated memory CD4 T cells and CD8 T cells（P=0.049） in the tumor microenvironment（TME） had a better prognosis.Conclusion CYTL1, PMP2 genes are associated with patient prognosis and could be potential targets for future treatment; the expression of PMP2, CYTL1 gene in the tumor microenvironment was negatively correlated with T cell follicular helper, CD8 T cells, and positively correlated with resting memory CD4 T cells, naive B cells. Activated memory CD4 T cells. CD8 T cells may be playing the protective factors in patients with bladder tumors.