Machine learning for structure learning of large dynamic systems in health and disease
Dynamic processes are increasingly studied by high-dimensional single-cell snapshots, complementing efforts based on time-lapse measurements. Such measurements enable us to evaluate cell population distributions and their evolution over time. However, it is not trivial to map these distributions across time and to identify dynamically important cell states. To this end we develop sparse regression, probabilistic graphical modeling, supervised surrogate learning approaches and operator approximation approaches, as well as supervised pseudotime approaches and simulation based trajectory inference for single-cell RNA seq time series analysis, and apply these to reconstruct differentiation processes such as T cell differentiation in cancer and chronic infections.