Non-invasive detection of urothelial carcinoma through ensembled methylation profiles and fragmentomic patterns in urinary cell-free DNA.

e16604 Background: The detection of epigenetic markers in urinary cell-free DNA (ucfDNA) has shown promise for tumor identification. However, the reliance on changes in methylation patterns presents challenges in the early detection of urothelial carcinoma, particularly when the tumor DNA present in urine is scarce. The incorporation of fragmentomic characteristics could enhance the detection of tumor signals. Methods: This study presents a single-assay approach, PredicineEPIC, which is a genome-wide methylation assay capable of analyzing both methylation and fragmentomic patterns in ucfDNA. Our cohort included 187 subjects: 66 with malignancies, 52 with benign conditions, and 69 healthy controls, divided into training and validation groups at a 2:1 ratio. We used the XGBoost machine learning algorithm for model development, followed by logistic regression to create an ensemble classifier. The training set utilized leave-one-out cross-validation to mitigate sample size constraints and prevent overfitting, with the classifier's performance assessed in the validation set. Results: The classifier's performance was evaluated using beta values of differentially methylated regions, the count of abnormally methylated fragments, and fragmentomic patterns (fragment sizes and transcription factor accessibility), with sensitivities of 0.75, 0.9, and 0.75, and a specificity of 0.86, respectively. The integrated features model showed a notable increase in sensitivity to 0.92 without losing specificity compared to models based on single features. This model also demonstrated precision on par with other methylation-based diagnostics, particularly in detecting early-stage tumors. Conclusions: Our study underscores an approach that differentiates malignant from benign lesions using a combination of omic features, enhancing the potential and precision of non-invasive liquid biopsy methods for early urothelial carcinoma detection.