Integrating of Machine Learning to Characterize Tumor-Infiltrating Immune Cell-Related Signature for Enhancing Prognosis for Glioma Patients.

Abstract

BACKGROUDS: Immunoblockade therapy based on PD-1 checkpoint has shown remarkable progress in various tumors, but its effectiveness in glioma patients is still lacking. Thus, it is in urgent need to uncover an ideal signature for glioma. METHODS: Five cohorts comprised 1973 patients from The Cancer Genome Atlas Program (TCGA), Chinese Glioma Genome Atlas (CGGA), and the Gene Expression Omnibus (GEO) database were included in the present study. By performing consensus clustering, limma and survival analysis, 42 prognostic genes were screened. Subsequently, a consensus immune cell infiltration-related signature (IRS) was developed using a 10-fold cross-validation framework with 101 combinations of 10 machine-learning algorithms, and the predictive performance of IRS was comprehensively analyzed. Ultimately, we evaluated the response of distinct risk subgroups to screen candidate drugs designed to address specific risk factors in the backgrounds of personalized medicine. RESULTS: Three molecular subtypes with distinct immune status and survival outcome were identified through consensus clustering analysis. By performing machine learning analysis on the common differentially expressed genes (DEGs), a consensus IRS was developed by the combined StepCox[both] + SuperPC algorithm. The IRS demonstrated high accuracy and robustness performance across multiple cohorts including TCGA, CGGA-693, CGGA325, GSE16011 and GSE43378. Moreover, the IRS could independently predict the survival outcome regardless of the impact of other clinical variables. Compared to the low IRS group, patients in the high IRS group are more sensitive to the chemotherapy drugs, while low IRS group patients may receive benefits from immunotherapy. Additionally, several candidate drugs were screened from multiple databases for poor survival outcome patients. Notably, a novel biomarker, C1QB, from the IRS, was highly expressed in glioma tissue and promotes progression in U87 cells by enhancing proliferative and migratory capacities while inhibiting apoptosis. CONCLUSIONS: The IRS could accurately predict glioma patients survival and may contribute to the development of personalized therapy. Moreover, as a key gene in IRS, over expression of C1QB could significantly enhance glioma cell viability.