Maximizing predictability of a bottom-up complex multi-scale model through systematic validation and multi-objective multi-level optimization

Abstract

Computational models are mathematical representations meant to replicate the biological system they represent, as well as provide insights and predict the system's dynamics in response to changing conditions. In a bottom-up modeling approach, a multitude of models may be compounded to represent more complex higher level biological systems. However, guaranteeing the validity and predictability of the compounded ensemble may become increasingly challenging as more components are integrated. We herein present a sequential and iterative method to maximize predictability of a complex multiscale model. We have successfully developed a multiscale modeling platform comprised of mechanisms ranging from the biomolecular level to multi-cellular networks. To maintain a high level of predictability of the global platform, we introduce a systematic approach to not only validate all models independently, but also verify the validity of compounded models as additional information becomes available at higher levels of complexity. Iterative and systematic application of these validation steps at increasing levels of complexity is intended to maximize the predictive power of the platform, making it a powerful tool to study the impacts of low-levels modifications (pathologies, drugs, etc.) on higher functional levels. The work presented lays down the rationale of the approach, the open design implementation and results.