169P - Urine tumor DNA fragmentation profiles provided a novel biomarker for early detection of prostate cancer

Background

Prostate Cancer is the second most common malignancy worldwide in male. Serum total PSA (tPSA) testing has led to the identification of patients with indolent prostate cancer, and inevitably over-treatment has become a concern. Urine tumor (utDNA) provides new approach for non-invasive early detection of prostate cancer.

Methods

Urine samples were collected from patients scheduled to undergo a prostate biopsy or from healthy volunteers. utDNA fragmentation profiles were analyzed by using low coverage whole genome sequencing. A multi-dimension machine learning algorithm combined large copy number variants (CNV), motifs and fragments distribution were developed by a customized bioinformatics workflow, urine tumor DNA multi-dimensional bioinformatic predictor (utLIFE). The utLIFE-PC score was calculated to improve the diagnostic value for prostate cancer by training and tested on a retrospective cohort, including case group with prostate cancer (PCa, n=30), control group with healthy individuals (n=24) and benign prostate hyperplasia (BPH, n=16).

Results

First, we calculated classification performance on single dimension by a 7:3 training:test set with 10-k fold across-validation. The large CNV feather could classify PCa from non-tumor controls with AUC 0.815, sensitivity 66.7% and specificity 88.2%. And the motif feather showed a better performance than large CNV with AUC 0.921, sensitivity 83.3% and specificity 94.1%. While, the AUC achieved 0.942(sensitivity 86.7% and specificity 84.3%) in the single analysis of fragment distribution. Based on these results, we developed a multi-dimension machine learning model, utLIFE-PC. The utLIFE-PC model showed a superior performance than any single dimension feather, with AUC 0.944, sensitivity 90.0% and specificity 92.2%.

Conclusions

The utLIFE-PC model showed high sensitivity and specificity in prostate cancer diagnosis when classifying PCa from healthy individuals and BPH. More importantly, urine samples could collect at home, and multi-dimension analysis was done with low coverage WGS. In breif, the utLIFE-PC model was superior to improve the prostate cancer diagnostic process and decrease unnecessary repeat biopsies with a easily and cheaply way.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.