Computationally efficient inference methods for biochemical reaction networks

Stochastic methods for simulating biochemical reaction networks often provide a more realistic description of biological processes over deterministic counterparts when small populations of certain chemical species are relevant, or when the system dynamics depend crucially on the presence of noise. Computationally expensive numerical methods are often required to analyse model behaviour due to the intractability of analytic solutions. The computational requirements are significantly compounded when the statistical inference of reaction rate parameters is considered. This computational bottleneck severely limits the number of unknown reaction rates that can be inferred. We investigate new techniques for the parameter inference problem with the aim of minimising computational overheads without compromising accuracy of the parameter estimates. We demonstrate our approach using a variety of discrete-state,continuous-time Markov models relevant to various applications in systems biology.