Enzymes can be used to catalyse reactions with splendid control over specificity and selectivity. However, their high degree of specialisation sometimes narrows down their application potential or complicates their engineering towards different functions. There are indications that ancestral enzymes used to be more promiscuous and more mutationally flexible, and therefore, they may hold great biotechnological appeal. Sadly, not much is currently known about which ancestral states are useful and how they can be applied for designing interesting novel biocatalysts. This project aims to answer those questions by reviving various ancient enzymes from a functionally diverse family of glycoside phosphorylases. Furthermore, alternative combinations of mutational scenarios will be made in order to explore a large range of possibilities from which natural evolution sampled merely a few options. Machine learning techniques will be used to guide the navigation of the entire historical sequence space, allowing us to predict enzyme variants that are better starting points for enzyme engineering, or that possess more beneficial traits as such.