CANCER RECURRENCE PREDICTION WITH MACHINE LEARNING: NOT JUST THE CELL PROLIFERATION RATE

Zohar Segal ¹ rahul Valiya Veettil ² noa Eyal-Altman ² mark Last ^2,3 Eitan Rubin ²

¹jusidman Science Center for Youth, Ben Gurion University of the Negev, Israel
²The Shraga Segal Dept. of Microbiology, Immunology and Genetics, Ben-Gurion University of the Negev, Israel
³department of Software and Information Systems Engineering, Ben-Gurion University of the Negev, Israel

Gene patterns reported to predict overall survival of breast cancer patients have been shown not to lose their predictivity upon normalization of the gene expression for cell division rate. In this work, we tested the hypothesis that recurrence prediction using machine learning also captures cell division rate rather than specific properties of the tumors that are more likely to recur. Comparing the ability of the XGBoost classification algorithm, we found that recurrence prediction accuracy is not affected by normalization for cell division rate. These results allow us to reject this hypothesis. By rejecting this hypothesis we support the hypothesis that cancer recurrence prediction captures signals that are not directly related to tumors’ cell proliferation rate.