Controlling the stability of vaccines thanks to modeling and prediction is a subject of significant interest in immunology and appeals to the focus of the World Health Organization (WHO) which provides guidelines on the subject. The first aim of stability prediction is to secure vaccine conformity at the time of injection, after whether an uneventful shelf-life. The matter particularly concerns pharmaceutical companies such as Sanofi Pasteur. Among the team Biologics Design & Generation of the Research and Development platform at Sanofi Pasteur, a Matlab-based software called Simstab was developed to adapt mathematical models to experimental data and therefore predict vaccine stability.This thesis describes the development of a method of Design of Simulated Experiments to improve the modeling of vaccine stability already given by the existing Simstab. An algorithm based on a Bayesian approach was developed to minimize the variance of a prediction, by suggesting new experimental measurements to implement. The software was applied on a influenza vaccine case study as proof of concept.A simple cumulative approach was able to suggest a few measurements to reduce the variance by 77%. If the measurements were constrained to 3 months of vaccine shelf-life, the results reached 58% of reduction. A more complex genetic algorithm method showed similar results in terms of experimental measurements suggested, but with less impact on the variance. The results imply that using high temperature experiments, and using repetition of measurements are the two factors which most impact the reduction of the variance of the prediction at 1 year of shelf-life and 5°C.
Jeanne-Marie Chevalier (Sun,) studied this question.