Work

Machine learning-enhanced discovery of a basement membrane-related gene signature in glioblastoma via single-cell and Spatial transcriptomics

Year: 2025

Type: article

Abstract: The complex invasiveness and heterogeneity of glioblastoma multiforme (GBM) hinder the complete eradication of the tumor. The invasion of the basement membrane (BM) occurs before the spread to the meninges and the metastasis of glioma cells, increasing the recurrence rate of the disease, leading to poor patient prognosis. In this study, four cohorts of GBM patients were aggregated to confirm the BM candidate genes through binary classification machine learning (ML). We delineated their features and developed a novel BM related genes (BMRGs) risk model through ML combinations, and compared it with previous models. Concurrently, the immune attributes of high and low-risk groups were analyzed using the immune infiltration algorithm encompassing CIBERSORT, xCell, and ssGSEA. Novel immunotherapeutic and chemotherapeutic strategies were proposed based on the assessment of immune treatment evaluation tools, TIDE and IRnet, and drug sensitivity analysis. Lastly, single-cell and spatial transcriptomics were utilized to unearth the properties of tumor cells with high BMRGs expression which were then verified via independent cohorts incorporating 42 samples from 17 patients. In accordance with ML combinations, four BMRGs provided the foundation for the predictive model, serving to categorize patients into groups of high and low risk. It was observed that increased risk scores correlated with diminished overall survival rates. Immune infiltration analysis revealed that high-risk group was characterized by a heightened infiltration of M2-like macrophages and a decrease in CD8 T cells. Concurrently, it was noted that high-risk group had reduced immunotherapy benefit scores. A drug sensitivity analysis revealed that high-risk group exhibited greater sensitivity towards five drugs. Furthermore, single-cell RNA sequencing (scRNA-seq) analyses discerned tumor cells with a high BMRGs score demonstrated more effective intercellular communication. The concluding observations revealed that BMRGs have enhanced expression and significant co-localization within GBM samples. This study elucidates a direct molecular correlation between GBM and BM, establishes a BMRGs signature risk model, and explores their interconnectivity. The findings offer pivotal insights into the clinical outcomes, therapeutic responses, and potential treatment strategies for GBM. Machine learning combinations depicted the characteristics of BM-related genes and established a BMRGs risk model. The BMRGs signature model shows good accuracy in the prognosis prediction of GBM patients. Patients in the high-risk group demonstrate reduced sensitivity to immunotherapy and heightened sensitivity to selected chemotherapy drugs. GBM cells with higher BMRGs expression have stronger communication functions. Spatial transcriptomics and independent cohort validation revealed that FOS regulates the expression of ADM, PLOD3 and AEBP1.

Source: Journal of Translational Medicine

DOI: 10.1186/s12967-025-06918-0