Amine transaminases (ATAs) are powerful biocatalysts for the stereoselective synthesis of chiral amines. Machine learning provides a promising approach for protein engineering, but activity prediction models for ATAs remain elusive due to the difficulty of obtaining high‐quality training data. Thus, we first created variants of the ATA from Ruegeria sp. (3FCR) with improved catalytic activity (up to 2000‐fold) as well as reversed stereoselectivity by a structure‐dependent rational design and collected a high‐quality dataset in this process. Subsequently, we designed a modified one‐hot code to describe steric and electronic effects of substrates and residues within ATAs. Finally, we built a gradient boosting regression tree predictor for catalytic activity and stereoselectivity, and applied this for the data‐driven design of optimized variants which then showed improved activity (up to 3‐fold compared to the best variants previously identified). We also demonstrated that the model can predict the catalytic activity for ATA variants of another origin by retraining with a small set of additional data.
Protein engineering is essential for altering the substrate scope, catalytic activity and selectivity of enzymes for applications in biocatalysis. However, traditional approaches, such as directed evolution and rational design, encounter the challenge in dealing with the experimental screening process of a large protein mutation space. Machine learning methods allow the approximation of protein fitness landscapes and the identification of catalytic patterns using limited experimental data, thus providing a new avenue to guide protein engineering campaigns. In this concept article, we review machine learning models that have been developed to assess enzyme‐substrate‐catalysis performance relationships aiming to improve enzymes through data‐driven protein engineering. Furthermore, we prospect the future development of this field to provide additional strategies and tools for achieving desired activities and selectivities.
More than 30 years ago, Diffusion-Limited Aggregation (DLA) has been studied as mechanism to generate structures sharing similarities with real-world cities. Recently, a stochastic gravitation model (SGM) has been proposed for the same purpose but representing a completely different mechanism. Both approaches have advantages and disadvantages, while e.g. the dendrites emerging via DLA are visually very different from real-world cities, the SGM does not preserve undeveloped locations close to the city center. Here we propose a unification of both mechanisms, i.e. a particle moves randomly and turns into an urban site with a probability that depends on the proximity to already developed sites. We study the cluster size distributions of the structures generated by both models and find that SGM generates more balanced distributions. We also propose a way to assess to which extent the largest cluster is a primate city and find that in both models, beyond certain parameter value, the size of the largest cluster becomes inconsistent with being drawn from the same distribution of remaining clusters.
Amintransaminasen (ATAs) sind leistungsstarke Biokatalysatoren für die stereoselektive Synthese chiraler Amine. Machine learning ist ein vielversprechender Ansatz für das Protein‐Engineering, aber Modelle zur Aktivitätsvorhersage für ATAs sind aufgrund der Schwierigkeit, qualitativ hochwertige Trainingsdaten zu erhalten, schwer zu finden. Aus diesem Grund haben wir zunächst Varianten der ATA aus Ruegeria sp. (3FCR) mit verbesserter katalytischer Aktivität (bis zu 2000‐fach) sowie umgekehrter Stereoselektivität durch strukturabhängiges rationales Design entwickelt und dabei einen hochwertigen Datensatz gesammelt. Anschließend entwarfen wir einen modifizierten One‐Hot‐Code, um sterische und elektronische Effekte von Substraten und Aminosäureresten innerhalb von ATAs zu beschreiben. Schließlich erstellten wir einen Gradient Boosting Regression Tree Prädiktor zur Vorhersage der katalytischen Aktivität und Stereoselektivität und wendeten diesen für das datengesteuerte Design optimierter Varianten an, die dann eine verbesserte Aktivität zeigten (bis zum Dreifachen im Vergleich zu den besten zuvor identifizierten Varianten). Wir haben zudem gezeigt, dass das Modell die katalytische Aktivität für ATA‐Varianten eines anderen Enzyms vorhersagen kann, indem es mit einem kleinen Satz zusätzlicher Daten neu trainiert wurde.