38P - Machine learning-based pathomics model to predict the infiltration of Treg and prognosis in IDH-wt GBM

Background

Regulatory T cells (Treg) play a crucial role as prognostic factors and intervention targets for isocitrate dehydrogenase (IDH)-wild type (wt) glioblastoma (GBM). The study aimed to construct a model predicting the Treg infiltration in IDH-wt GBM patients using pathomics techniques and explore related biological processes.

Methods

Flow cytometry was used to detect the proportion of Treg in orthotopic mouse glioblastoma brain tissue. Clinical data of IDH-wt GBM patients from the TCGA database were analyzed retrospectively. Features were extracted from HE-stained biopsy sections using the Pyradiomics package. The pathomics model was constructed using the Gradient Boosting Machine algorithm after performing feature selection with mRMR and Relief algorithms. Cox proportional hazard regression analysis was employed to access the association between pathomics score (PS) and overall survival (OS). Transcriptomic data were analyzed through GSEA set enrichment, differential gene expression, and correlation analyses.

Results

The study established a pathomics model with 3 pathomics features that effectively predicted Treg infiltration (Training set: AUC=0.807; Validation set: AUC=0.735). PS positively correlated with high Treg expression. Survival analysis indicated that patients with high PS had significantly lower OS than the low PS group (median survival time: 12.0 vs 14.4 months; p=0.009). Multivariate COX analysis revealed that high PS expression independently served as a prognostic risk factor for IDH-wt GBM patients (HR, 2.16; 95% CI, 1.269-3.677; p=0.005). Subgroup analysis showed that high PS was a risk factor for OS in patients receiving chemotherapy (HR, 2.232; 95% CI, 1.309-3.905; p=0.003) or radiotherapy (HR, 1.882; 95% CI, 1.161-3.049; p=0.01).GSEA enrichment analysis revealed significant associations of PS with the NOTCH, IL-6/JAK/STAT3 signaling pathways. High PS significantly correlated with elevated RAD50 expression.

Conclusions

The developed pathomics model, based on machine learning algorithms, can noninvasively predict Treg infiltration and prognosis in IDH-wt GBM patients. The biological processes related to the model may involve RAD50 and pathways, including NOTCH, IL-6/JAK/STAT3.

Legal entity responsible for the study

The authors.

Funding

The Joint Funds for the Innovation of Science and Technology, Fujian Province (No.2021Y9301).

Disclosure

All authors have declared no conflicts of interest.