Skip to main content
. 2025 Jan 20;16:865. doi: 10.1038/s41467-024-55399-0

Fig. 5. ML-guided engineering of distinct amide synthetases for the biosynthesis of a broad panel of small-molecule pharmaceuticals.

Fig. 5

A The strategy we used for machine learning-guided protein engineering of McbA is shown. First, we identified non-native reactions that wt-McbA can catalyze and prioritized those that produce valuable small-molecule pharmaceuticals. Second, an HSS of 64 residues is used to down-select residues that positively impact activity. Third, an augmented ridge regression model is trained on data from the HSS, and ML predictions are experimentally tested. B–G Comparison of the highest activity predicted variant for a panel of small-molecule pharmaceuticals compared to wt-McbA and an authentic standard. Enzyme concentration was normalized to 0.5 mg/mL (~9 µM) and products were analyzed by RP-HPLC. The fold-increase in yield observed compares wt-McbA to ML-McbA (n = 3). Representative HPLC traces of product (red), acid substrate (purple), and adenylated acid (orange) for each reaction are shown. Traces are taken from at least three independent experiments (n = 3).