Marios Tsolekas, "Finding the best configuration for biological simulations at PhysiBoSS", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2021
https://doi.org/10.26233/heallink.tuc.89720
Biological simulations for simulating large cell populations have become anecessary tool in modern science and the software that has been developedwith that goal in mind is complex, having to emulate not only howneighboring cells interact with each other, but also how external stimuliand environmental factors affect the population as a whole and how itaffects groups of it. It is of no surprise then, that these simulations have aconsiderable execution time, which makes it time consuming to exhaustivelyrun those simulations for different configurations in search of an optimalresult.The system presented in this thesis ventures to remedy that problemby leveraging Bayesian optimization to stochastically find that optimalconfiguration, for PhysiBoSS’s biological simulations, that minimizes thenumber of alive cancer cells. By treating simulations as black box functionsand modeling them based on samples selected by an acquisition function, itis possible to rapidly converge to an optimal configuration that correspondsto the desired global optima using only a small number of simulations, forexample in a search space of a few thousand data points this process wouldrequire less than fifty samples.