Development of a computational pipeline for the calibration of treatment response in an ovarian cancer digital twin

Treatment of High-grade serous ovarian cancer (HGSOC) is often ineffective due to frequent late-stage diagnosis and development of resistance to therapy. Timely selection of the most effective combination of drugs for each patient would improve outcomes. A computational simulator capable of recapitulating cell response to treatment in ovarian cancer was built for in silico studies. This simulator (named ALISON) is a digital twin, a virtual representation of a biological system that can be calibrated to recapitulate the behavior of individual cell lines and patients. ALISON focuses on simulating a virtual representation of the lining of the peritoneum (the omentum), as HGSOC primarily spreads within the abdominal cavity.Cell behavior is described through a set of specific states that each cell can assume and programmable user-defined rules that formalize allowed behaviors and corresponding likelihood of occurrence. These rules are defined as a system of equations whose parameters need to be calibrated to ensure the simulated behavior matches the experimental one as closely as possible. Parameters are divided into basal behavior and drug-response parameters. In this master’s thesis project, the aim was to calibrate model parameters of 124 drugs administered to four ovarian cancer cell lines previously calibrated under basal conditions to achieve the 50% of cells inhibition (IC50), using the GDSC (Sanger) for drug mechanisms and effective concentrations as reference.Calibration was performed by simulating cell lines multiple times generating a set of random parameters with Latin Hypercube Sampling for each drug. This enabled the acquisition of a sufficiently large dataset to train a machine learning algorithm to accelerate parameter optimization using a genetic algorithm. This tool improved the efficiency of the parameter calibration procedure.