This thesis uses a variety of statistical and machine learning techniques to provide new insight into astrochemical processes. Astrochemistry is the study of chemistry in the universe. Due to the highly non-linear nature of a variety of competing factors, it is often difficult to understand the impact of any individual parameter on the abundance of molecules of interest. It is for this reason we present a number of techniques that provide insight.

Chapter 2 is a chemical modelling study that considers the sensitivity of a glycine chemical network to the addition of two H2 addition reactions across a number of physical environments. This work considers the concept of a “hydrogen economy” within the context of chemical reaction networks and demonstrates that H2 decreases the abundance of glycine, one of the simplest amino acids, as well as its precursors.

Chapter 3 considers a methodology that involves utilising the topology of a chemical network in order to accelerate the Bayesian inference problem by reducing the dimensionality of the parameters to be inferred at once. We demonstrate that a network can be simplified as well as split into smaller pieces for the inference problem by using a toy network.

Chapter 4 considers how the dimensionality can be simplified by exploiting the physics of the underlying chemical reaction mechanisms. We do this by realising that the most pertinent reaction rate parameter is the binding energy of the more mobile species. This significantly reduces the dimensionality of the problem we have to solve.

Chapter 5 builds on the work done in Chapters 3 and 4. The MOPED algorithm is utilised to identify which species should be prioritised for detection in order to reduce the variance of our binding energy posterior distributions.

Chapter 6 introduces the use of machine learning interpretability to provide better insights into the relationships between the physical input parameters of a chemical code and the final abundances of various species. By identifying the relative importance of various parameters and quantifying this, we make qualitative comparisons to observations and demonstrate good agreement.

Chapter 7 uses the same methods as in Chapters 4, 5 and 6 in light of new JWST observations. The relationship between binding energies and the abundances of species is also explored using machine learning interpretability techniques.