Machine Learning Guided Rational Design of a Non-Heme Iron-Based Lysine Dioxygenase Improves its Total Turnover...

Area of Science:

• Organic Chemistry
• Biocatalysis
• Protein Engineering
• Machine Learning

Background:

• Selective C-H functionalization is a significant challenge in synthetic organic chemistry.
• Biocatalysts offer powerful solutions for complex chemical transformations.
• Traditional enzyme engineering relies on directed evolution or rational design, which can be time-consuming.

Purpose of the Study:

• To integrate machine learning with molecular dynamics for efficient biocatalyst engineering.
• To rationally design improved variants of non-heme iron-dependent lysine dioxygenase (LDO).
• To reduce the experimental burden in identifying beneficial enzyme mutations.

Main Methods:

• Utilized MutComputeX, a structure-based self-supervised machine learning framework.
• Employed classical molecular dynamics simulations to predict and down-select mutations.
• Applied these computational methods for the rational design of LDO mutants.

Main Results:

• The combined computational approach consistently yielded functional LDO mutants.
• Engineered single mutants showed up to twofold higher expression yields compared to wild-type (WT).
• A pentamutant variant (LPNYI LDO) exhibited a 40% increase in total turnover number (TTN) over WT LDO.

Conclusions:

• Synergistic application of machine learning and molecular dynamics streamlines protein engineering.
• This low-barrier approach enables efficient discovery of improved biocatalysts.
• The developed strategy facilitates the optimization of enzymes for challenging synthetic transformations.