Experimental and computational exploration of enzyme sequence space

Sammanfattning: Millions of enzymes with desirable features or new exciting activities can be found in organisms occupying diverse niches all around the earth. However, enzyme studies tend to be biased towards characterisation of representatives from eukaryotes, model organisms, or disease-causing bacteria. As such, a large number of enzymes still remains underexplored. The so-called sequence space of proteins - all possible protein sequences - is even greater when we include not only natural sequences, but also the ones designed by human or artificial intelligence. This thesis explores various reasons, approaches, and outcomes of investigation of large enzymatic sequence spaces.  In the first part of my work, I focused on investigation of a natural sequence space of oxidases using a high-throughput activity profiling platform. A functional screen of an industrially important class of enzymes, S-2-hydroxyacid oxidases (EC 1.1.3.15), revealed that nearly 80% of the class is misannotated. Further exploration of annotations to public databases indicated that similar errors of annotations can be found in other enzyme classes. A broader activity profiling of 1.1.3.x oxidases resulted in the discovery of two novel microbial enzymes: N-acetyl-hexosamine oxidase, and a novel type of long-chain alcohol oxidase.  Natural enzymes often need to be improved in order to be industrially applied, for example to become more stable, or accept non-natural substrates. A novel, and constantly developing, approach for enzyme design involves the use of machine learning (ML) tools. Second part of my work focused on screening an enzyme sequence space designed by generative adversarial networks. Our work proved that ML methods can generate fully functional enzymes that mimic sequences present in nature. Enzyme assays are necessary to get a full understanding of how enzymes work. Traditional kinetic assays are time- and reagent-consuming and as a result a limited number of variants and conditions are being tested for each target. In my final work I described a novel approach for enzyme kinetic studies, by adaptation of a microfluidic qPCR device.