A miRNA-BASED PREDICTOR OF PROGNOSIS AND THERAPEUTIC RESPONSE FOR HIGH-GRADE SEROUS EPITHELIAL OVARIAN CANCER USING MACHINE LEARNING

Vol 3, 2022 - 155259
MS - Masters student
Favorite this paper
How to cite this paper?
Abstract

INTRODUCTION AND OBJECTIVES: Ovarian cancer (OvC) is one of the neoplasms with the highest incidence among women worldwide, with significantly increased mortality. The vast majority of patients are diagnosed in advanced stages of the disease since early stages present unspecific symptoms and inaccurate diagnoses. Currently, not many clinical biomarkers are available to aid in the diagnosis or predict the prognosis of HGSOC. The application of omics approaches on non-invasive samples, such as liquid biopsies, combined with artificial intelligence techniques shows promise, not only for identifying effective biomarkers for prognosis and diagnosis of OvC, but also for decision making. Therefore, our goal is to build a predictor of prognosis for patients diagnosed with ovarian cancer, as well as to identify new biomarkers. MATERIAL AND METHODS: Prognostic groups were defined based on the number of years of overall survival and their vital status: Patients with less than 3 years of survival and with Dead status were allocated as the group with poor prognosis and patients with ≥ 3 years were considered as a good prognosis. Based on the miRNA expression data from samples of patients diagnosed with ovarian cancer obtained from the database of The Cancer Genome Atlas (TCGA) project, we applied variable selection methods (FCBF, Cox Univariate Regression, and ElasticNet) to identify relevant predictors. Subsequently, samples were subject to a ten-fold split (outer cross-validation), wherein each 10% split was used as a held-out set, with the remaining 90% used as a training set. Random Forest (RF) classifiers were trained with hyperparameter optimization performed through a grid search with ten-fold internal cross-validation, leading to a total of ten separate AUC (Area Under Curve) values. The mean of these ten values was calculated as a summary AUC metric. RESULTS AND CONCLUSION: After applying the variable selection methods, 47 miRNAs were selected as important for the clinical outcome of patients with OvC. Regarding the performance of the models on the test data set, RF showed excellent metrics: 0.848 AUC (Area Under the Curve); 0.761 sensitivity; 0.771 precision, and with an accuracy of 0.763. Among the miRNAs with greater predictive potential, miRNA-205-5p was considered an excellent predictor by the RF model to classify patients with good or poor prognosis regarding survival. Published studies show that this miRNA has a mechanism of action dependent on tumor cell-derived exosomes and has the ability to regulate angiogenesis in the tumor, thus being a potential therapeutic target for OvC. Furthermore, independently evaluating the miRNAs indicated by the RF model as having the highest predictive potential, miRNA-143-3p showed an AUC of 0.700. Through in vitro experimental studies, authors have already demonstrated that this miRNA acts as a tumor suppressor in ovarian cancer cell lines.

Share your ideas or questions with the authors!

Did you know that the greatest stimulus in scientific and cultural development is curiosity? Leave your questions or suggestions to the author!

Sign in to interact

Have a question or suggestion? Share your feedback with the authors!

Institutions
  • 1 Bioinformatics and Computational Biology Lab, Division of Experimental and Translational Research, Brazilian National Cancer Institute, RJ, Brazil
  • 2 Big Data and Predictive Analysis in Health Lab, School of Public Health, University of São Paulo (USP), SP, Brazil
Track
  • 1. Bioinformatics
Keywords
miRNA
Machine Learning
Prognosis
liquid biopsies