Connecting chemical and protein sequence space to predict biocatalytic reactions

Abstract

The application of biocatalysis in synthesis has the potential to offer streamlined routes towards target molecules1, tunable catalyst-controlled selectivity2, as well as processes with improved sustainability3. Despite these advantages, biocatalysis is often a high-risk strategy to implement, as identifying an enzyme capable of performing chemistry on a specific intermediate required for a synthesis can be a roadblock that requires extensive screening of enzymes and protein engineering to overcome4. Strategies for predicting which enzyme and small molecule are compatible have been hindered by the lack of well-studied biocatalytic reaction datasets5. The underexploration of connections between chemical and protein sequence space constrains navigation between these two landscapes. Here we report a two-phase effort relying on high-throughput experimentation to populate connections between productive substrate and enzyme pairs and the subsequent development of a tool, CATNIP, for predicting compatible α-ketoglutarate (α-KG)/Fe(ii)-dependent enzymes for a given substrate or, conversely, for ranking potential substrates for a given α-KG/Fe(ii)-dependent enzyme sequence. We anticipate that our approach can be readily expanded to further enzyme and transformation classes and will derisk the investigation and application of biocatalytic methods. A two-phase machine-learning-based tool making use of high-throughput experimentation is introduced to examine the connections between chemical and protein sequence space and predict productive biocatalytic reactions among substrate and enzyme pairs.