MULTIMODE: Methodologies and Tools for the Analysis and Design of Multimodal Stochastic Systems

• Applicant: Professorin Dr. Verena Wolf
• Subject Area: Data Management, Data-Intensive Systems, Computer Science Methods in Business Informatics
• Term: from 2018 to 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 391984329

In this project we will devise novel computational methods in the area of modelling and simulation and apply them to problems in systems biology. The project focuses on the development of algorithms for the analysis of quantitative stochastic models with discrete state spaces having a multimodal behaviour. Such models are widely used to describe probabilistic decisions in cellular processes where stochasticity is a driver of phenotypic diversity even among monoclonal cells. Their simulation and calibration is an indispensable part of the research cycle that connects computational modelling and wet-lab experiments, but is also useful for model-based complex systems engineering in general. The main goal is to develop efficient methods to estimate the steady state distribution of stochastic models and their parameters. We will focus on approximation algorithms that are tailored to systems with multimodal equilibrium behaviour, i.e., systems with multiple attracting regions. Instead of relying on sampling techniques or on direct numerical approaches, we will propose methods that are mostly based on equations for the statistical moments of the steady state distribution. Current research in the area has either concentrated on general theories (such as mean-field theory and moment closure techniques) for non-equilibrium processes or on crafting solutions for specific systems. Here, our breakthrough idea is to tackle the problem of steady state estimation for large multimodal systems by decomposing it into the localization of modes (that is, attracting regions), the reconstruction of mode conditional probabilities, and the computation of the relative strength of each mode. This decomposition will strongly increase the performance of solution algorithms and will ensure that the computation times are independent of maximum molecular counts.We will then leverage these novel approximation algorithms to tackle two related fundamental challenges in the area: model calibration from experimental data and model synthesis, which are both only feasible if efficient solution methods are available. The methodological results of this project will contribute novel techniques to the field of modelling and simulation of complex systems and, with novel software tools, allow to increase our understanding of probabilistic cellular behaviour.

• DFG Programme: Research Grants