Prediction of the clinical outcomes of prostate cancer remains a challenge. Recently, we discovered a panel of 8 fusion genes that occurred in aggressive prostate cancer. In order to make the fusion gene test clinically ready as a predictor, we have modified to test into a semi-quantitative Taqman QRT-PCR. In the funding period, Two hundred seventy-one prostate cancer samples with clinical follow-up were collected from University of Pittsburgh Medical Center. In addition, 194 prostate cancer samples from University of Wisconsin, Madison and 108 prostate cancer samples from Stanford University were collected. Taqman QRT-PCRs were performed on these samples. Significant numbers of samples were found positive for some of these fusion genes. The expression of MAN2A1-FER, SLC45A2-AMACR, MTOR-TP53BP1 fusions are associated with prostate cancer recurrence in the UPMC cohort. Cross-validation showed that fusion gene model predicts up to 91% clinical outcomes of prostate cancer accurately. When cohorts of UPMC, Stanford and Wisconsin were combined, the accuracy is 74%. The combination of fusion with Gleason appeared to improve the overall accuracy from 77% (Gleason) to 92% (Gleason+fusion) in the UPMC cohort, and from 71% (Gleason) to 82% (Gleason+fusion) when all three cohorts are combined. When fusion combined with both pathology stage and Gleason, the accuracy was improved a little further: 93% accuracy in the UPMC cohort and 83% when all three cohorts are combined. In summary, fusion transcript prediction model may have a role in prostate cancer prognosis prediction and guiding the management of prostate cancer patients.

Citation Format: Jian-Hua Luo, Silvia Liu, Yan-Ping Yu. Clinical outcome prediction of prostate cancer using machine learning based on fusion gene detection [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr LB150.