# Single-cell sequencing combined with machine learning to identify glioma biomarkers and therapeutic targets

## TL;DR

This study uses single-cell sequencing and machine learning to find biomarkers and therapeutic targets for glioblastoma multiforme.

## Contribution

A 16-gene signature and four key biomarkers for glioma prognosis and treatment were identified using machine learning and ubiquitination analysis.

## Key findings

- Three distinct cell differentiation stages were identified in glioma tissues using single-cell RNA-seq data.
- A 16-gene DRG signature was developed for predicting glioma patient survival, with 13 genes related to ubiquitination.
- ETV4 inhibition was shown to reduce glioma cell proliferation and invasion.

## Abstract

The purpose of this study is to utilize single-cell sequencing data to explore glioma heterogeneity and identify key biomarkers associated with glioblastoma multiforme (GBM) relapse using machine learning.

Single-cell sequencing and transcriptome data for gliomas were obtained from the GEO (GSE159416, GSE159605, and GSE186057) and TCGA databases. A prognostic model based on differentiation-related genes (DRGs) was constructed using weighted correlation network analysis, univariate Cox regression, and LASSO analysis. Key genes were identified using LASSO and SVM-RFE, with intersecting genes selected as the final set of key genes. Further analyses examined immune infiltration patterns and functional pathways. Importantly, we analyzed the relationship between prognostic-related genes and ubiquitination, and further characterized the characteristics of ubiquitination-related prognostic genes. In addition, we performed CCK-8 assays, colony formation, Transwell invasion assays, apoptosis assays to determine the role of ETV4 in glioma.

Examination of single-cell RNA-seq data from the GEO database revealed three distinct cell differentiation stages in glioma tissues. Marker genes for each of these cell states were combined to form DRGs. A 16-gene DRG signature was developed for predicting the survival of glioma patients. Machine learning identified four important genes with high AUCs in both training and test sets. Notably, 13 out of 16 genes in the DRG signature are ubiquitin-related, highlighting the involvement of ubiquitination in GBM. Moreover, we reported that inhibition of ETV4 attenuates cell proliferation and invasion in glioma cells.

Our prognostic model, based on the differentiation-related gene signatures, may be valuable for predicting prognosis and immunotherapy response in glioma patients. Characterizing these ubiquitination-associated features may elucidate the molecular mechanisms driving GBM progression and offer novel insights for its diagnosis and treatment. Additionally, machine learning identified four biomarkers with potential for aiding in the diagnosis and treatment of GBM.