The efficiency, sustainability and environmental friendliness of many industrial chemical reactions can be improved by using enzymes to catalyse the reaction. One important factor to be considered when selecting or designing an enzyme for a particular application is the temperature range at which the enzyme is most effective, but determining this experimentally is time-consuming and expensive.
The aim of this project is to use machine learning methods, together with novel 3D structural feature sets, to develop models that can predict the optimal catalytic temperature for enzymes. One possible use of these models is in the identification of thermozymes to be used in biocatalysis and biofuel production.
This question has been studied in previous projects (Li et al., 2019; Ulfenborg, 2020). We now wish to extend this work by adding many new features and experimenting with other machine learning methods in order to improve predictive performance.
This is an exploratory scientific project, with a lot of scope for creativity and innovation. You will get ideas iand suggestions from the supervisors, but you are encouraged to devise and implement your own features, and to experiment with advanced machine learning methods.
A course in machine learning. The course "Computational methods in bioinformatics" (Chalmers: TDA507, GU: DIT742) is recommended., but is not essential.
Martin Engqvist and Graham Kemp