Machine learning-based pathomics model to predict the infiltration of regulatory T cells and prognosis in isocitrate dehydrogenase-wild- type glioblastoma

Abstract

Abstract Purpose Regulatory T cells (Tregs) have been highlighted as prognostic factors in isocitrate dehydrogenase (IDH)-wild-type (wt) glioblastoma (GBM). However, conventional detection of Tregs with immunohistochemistry is limited for practical application in clinical settings. The aim of this study was to construct a pathomics model to predict Treg infiltration in IDH-wt GBM and explore the related biological processes. Methods Using the Pyradiomics package, pathomics features were extracted from hematoxylin and eosin-stained biopsy images of patients from The Cancer Genome Atlas. The proportion of Tregs was confirmed in orthotopic glioblastoma mouse model via flow cytometry. The pathomics model was constructed using a gradient-boosting machine-learning approach, and the pathomics score (PS) was determined with the minimal redundancy-maximal relevance and relief algorithms. Cox proportional hazard regression analysis was employed to access the association between PS and overall survival (OS). Transcriptomic data were analyzed through GSEA set enrichment, differential gene expression, and correlation analyses. Results PS was positively correlated with high Treg expression. Patients with a high PS had significantly worse overall survival than did those with a low PS. A high PS independently served as a prognostic risk factor for patients with IDH-wt GBM. Gene set enrichment analysis revealed significant associations between PS and the Notch and IL-6/JAK/STAT3 signaling pathways. A high PS was also significantly correlated with elevated RAD50 expression. Conclusion The developed pathomics model based on machine-learning algorithms can offer an alternative non-invasive method to predict Treg infiltration and prognosis in patients with IDH-wt GBM, further suggesting potential targets for immunotherapy.