Integrating transcriptomics, single-cell omics, and deep learning-based histopathological features to identify OLFML3 in bladder cancer.

BACKGROUND: Bladder cancer (BCa) represents the most common malignancy of the urinary system, characterized by a high recurrence rate, with approximately 61% of patients experiencing recurrence within 1 year post-surgery. Current monitoring methods, such as cystoscopy and urine cytology, are constrained by low sensitivity and patient discomfort. This study employed multi-omics data to investigate the role of OLFML3 in BCa recurrence, with the aim of enhancing the accuracy of recurrence prediction and improving clinical management.

METHODS: This study utilized RNA sequencing (RNA-seq) data and clinical information from The Cancer Genome Atlas (TCGA) to analyze patients with BCa, stratifying them into relapse and non-relapse groups. Weighted gene co-expression network analysis (WGCNA) was performed to identify gene modules associated with 1-year BCa recurrence. Subsequently, univariate Cox regression and least absolute shrinkage and selection operator (LASSO) Cox regression analyses were conducted to select eight prognostic genes, and a risk model was developed and validated in both TCGA and Gene Expression Omnibus (GEO) datasets. Additionally, single-cell RNA sequencing (scRNA-seq) data from Guangdong Provincial Second People's Hospital were analyzed to evaluate gene expression across high and low tumor stromal BCa subtypes and explore the relationship between the expression of these eight genes and clinical features. A deep learning model based on the ResNet50 architecture was developed to predict OLFML3 expression in hematoxylin and eosin-stained images. Statistical analysis was performed using R software (version 4.4.2), with significance set at P<0.05.

RESULTS: WGCNA identified gene modules associated with BCa recurrence, with the red module exhibiting a significantly positive correlation with recurrence status. Through univariate and LASSO Cox regression analyses, we selected eight prognosis-related genes. Kaplan-Meier survival analysis demonstrated that these genes effectively differentiated between high- and low-risk groups, with statistically significant survival differences observed in both TCGA and GEO datasets. Further Kaplan-Meier survival analysis of each of the eight genes indicated that high OLFML3 expression was associated with lower survival rates. scRNA-seq revealed differential expression of OLFML3 between high and low tumor stromal subtypes, suggesting that OLFML3 may be a key gene associated with 1-year BCa recurrence. High OLFML3 expression also correlated with BCa invasiveness, grade, and stage. Using a deep learning model based on BCa pathological features, we constructed a random forest (RF) model that successfully predicted high and low OLFML3 expression levels, providing novel insights for the clinical prognosis of BCa recurrence within 1 year.

CONCLUSIONS: Multi-omics approaches effectively identified the OLFML3 gene as a critical factor potentially associated with 1-year BCa recurrence. Furthermore, a deep learning model based on pathological features was developed to predict OLFML3 expression. Further research is warranted to elucidate its role in BCa recurrence within 1 year.