Introduction To implement precision cancer medicine in ovarian cancer, precise prediction of treatment response and identification of patients at high risk of disease recurrence are the first steps. Thus, we aimed to develop a next-generation RNA sequencing-based deep-learning model predicting chemoresistance risk in high-grade serous ovarian carcinoma (HGSCC).

Methods We conducted next-generation RNA sequencing on fresh-frozen, chemotherapy-naïve primary HGSOC tissues from 86 patients. Patients were randomly divided into training and test sets at a 1:1 ratio. In the model development phase, transcriptomic data from both the training set and The Cancer Genome Atlas HGSOC patients (n=419) were used. Using genes selected by the gene expression ratio analysis, we constructed and trained a deep neural network (DNN). Multiple DNN models were combined to build average ensemble models, which were further validated using the test set in the validation phase.

Results All patients in the study population received platinum-based combination chemotherapy: 15 and 71 were identified as chemoresistant and chemosensitive, respectively. Based on the gene expression ratio between chemoresistant and chemosensitive groups, we selected the top 70 genes with high expression ratios. Machine learning algorithms were applied to develop and train DNNs of the selected genes. Then, the five-fold average ensemble models were developed. Among the various ensemble models, the best model predicted chemoresistant cases with high accuracy (AUC, 0.925).

Conclusion/Implications We successfully developed next-generation RNA sequencing-based deep-learning models to predict chemoresistance risk after first-line platinum-based chemotherapy in HGSOC. These newly developed models would help the individualized management of HGSOC patients.