Oxidative stress reprograms benign prostatic hyperplasia microenvironments: insights from integrative multi-omics and machine learning

Comprehensive Summary

Chen et al. used machine learning and multiomics data to examine how oxidative stress shapes the microenvironment in benign prostatic hyperplasia (BPH). The analysis combined bulk RNA sequencing, single-cell RNA sequencing, and spatial transcriptomics data, along with experimental validation. Across datasets, the authors identified 499 differentially expressed genes and narrowed this down to 26 oxidative stress-related genes. Multiple machine learning methods (LASSO, random forest, SVM-RFE, and Boruta) identified three key genes: ACOX2, CTSB, and SERPINF1, all of which were upregulated in BPH. These genes showed strong diagnostic performance, with AUC values ranging from about 0.83 to 0.88, and a combined model reaching an AUC of 0.92. Oxidative stress levels were higher in BPH across most cell types, especially fibroblasts. ACOX2-positive fibroblasts were expanded in BPH and showed increased signaling activity and progression toward activated states. Experimental data supported this, showing that oxidative stress increased ACOX2 expression and led to higher cell proliferation and reduced apoptosis.

Outcomes and Implications

These results suggest that machine learning can help identify specific molecular drivers of BPH rather than just confirming general pathways like oxidative stress. Instead of relying on broad markers, this approach highlights a smaller set of genes that may better capture disease behavior. The expansion of ACOX2-positive fibroblasts also points to a possible mechanism linking oxidative stress to tissue remodeling and disease progression. This could help guide future biomarker development or targeted therapies. However, the study has clear limitations. Most findings are based on retrospective datasets and computational analysis, with limited validation in human cohorts. The machine learning models identify associations but do not prove causation, and drug predictions are based on simulations rather than real clinical testing. Overall, these findings are useful for generating hypotheses, but further validation is needed before applying them in clinical practice.