Statistical inference for intractable simulator-based models.
Likelihood-free inference and Approximate Bayesian Computation (ABC) are important emerging computational inference methods for complex statistical models in a wide range of applications ranging from cancer evolution to modeling of ecosystems. We work both on inference algorithm development and applications of the methods to a multitude of challenging problems in collaboration with biologists, epidemiologists and physicists.
Theory and applications of generalized graphical models.
Graphical statistical models are ubiquitous in science and engineering applications. We develop classes of graphical models that are context-aware and represent more flexible distributional assumptions. Bayesian methods for learning such models from data are also extensively considered. In our work these various model families are applied to predictive classification, causal inference and large-dimensional network analysis of genome sequence data.
Bayesian methods for DNA sequence analysis and population genomics for bacteria and viruses.
The explosion of population based DNA sequence data has enabled entirely novel perspectives on evolutionary epidemiology. We develop tools for large-scale genome data analysis and apply them in collaboration with leading groups in pathogen genomics to gain insight to evolutionary processes of relevance for human and animal health.