A machine learning-based workflow for transaminase selection

Area of Science:

• Biocatalysis and Enzyme Engineering
• Medicinal Chemistry and Drug Discovery
• Machine Learning in Chemistry

Background:

• Transaminases are valuable biocatalysts for synthesizing chiral amines, essential building blocks in medicinal chemistry.
• Selecting the optimal transaminase variant for specific substrates is challenging due to limited reactivity data.
• Current methods often require empirical screening or chiral preparative separations, which are inefficient.

Purpose of the Study:

• To develop predictive machine learning models for transaminase activity and selectivity.
• To create a high-throughput experimentation (HTE) dataset for training and validating machine learning models.
• To demonstrate the utility of these models in selecting appropriate transaminases for synthesizing relevant chiral amines.

Main Methods:

• Construction of a dataset comprising 336 transaminase reactions utilizing high-throughput experimentation (HTE).
• Development and application of machine learning (ML) models to predict enzyme performance (activity and selectivity).
• Validation of model predictions against baseline experimental results and prospective application to a held-out dataset.

Main Results:

• Machine learning models accurately predicted substrate conversions and top-k enzyme selection for selectivity.
• The developed models outperformed traditional baseline experimental approaches in predictive accuracy.
• Successful prospective application of the modeling workflow for selecting transaminases for cyclic ketone substrates.

Conclusions:

• Machine learning-guided enzyme selection significantly enhances the efficiency of chiral amine synthesis using transaminases.
• The developed HTE-ML workflow provides a powerful tool for accelerating drug discovery by optimizing biocatalytic routes.
• This approach alleviates the need for laborious chiral separations, offering a more sustainable and cost-effective synthesis strategy.