Unraveling bladder cancer-related circRNA biomarkers: a hybrid model combining deep learning and statistics

Circular RNAs (circRNAs), a class of gene expression regulators with closed loop structures, have been implicated in the pathogenesis of various diseases, including bladder cancer (BC). Their stability in and presence in body fluids suggest potential as diagnostic and prognostic biomarkers. This study employed an ensemble model integrating deep learning, feature selection, and statistical analysis to identify BC-associated circRNA biomarkers from RNA-sequencing data of 454 BC patients and 19 healthy controls. Utilizing the CIRCexplorer3 tool, we preprocessed the RNA-seq data, identifying 22,311 circRNAs and categorizing them based on expression patterns into three classes: exclusive to BC, exclusive to healthy controls, and present in both. Statistical methods were applied to the first class, while an integrated approach combining feature selection, feature extraction, and deep learning was employed for the third class. Our approach yielded 34 novel circRNA biomarkers, of which 17 have corresponding genes with verified associations with BC, validating the effectiveness of our method. The remaining 17 circRNAs represent potential novel BC circRNA biomarkers. The results underscore the efficacy of our computational strategy in advancing BC non-invasive biomarker discovery, potentially aiding early diagnosis and prognosis of BC.