Related Experiment Video
Updated: Jan 18, 2026
08:59
Defining Substrate Specificities for Lipase and Phospholipase Candidates
Published on: November 23, 2016
15.5K
Resolution of physics and deep learning-based protein engineering filters: A case study with a lipase for industrial
Spencer Gardiner1, Peter Dollinger2, Filip Kovacic2,3
1Department of Physics and Astronomy, Brigham Young University, Provo, Utah, United States of America.
Plos One
|September 12, 2025
Summary
Computational enzyme design struggles to rank subtle activity changes in Pseudomonas aeruginosa LipA. Current methods, including deep learning, cannot distinguish minor functional differences, highlighting the need for improved computational tools for biocatalyst optimization.
Area of Science:
- Enzyme engineering and computational biochemistry.
- Protein structure-function relationship analysis.
- Biocatalysis and industrial enzyme applications.
Background:
- Computational enzyme design is crucial for optimizing biocatalysts, particularly for non-natural substrates.
- Pseudomonas aeruginosa LipA, a lipase with a flexible lid, is a target for enhancing hydrolysis of the industrially relevant Roche ester.
- Existing computational tools face limitations in predicting subtle functional improvements in enzyme variants.
Purpose of the Study:
- To investigate and enhance the hydrolysis activity of Pseudomonas aeruginosa LipA towards Roche ester using computational enzyme design.
- To evaluate the effectiveness of a computational pipeline integrating molecular dynamics (MD) simulations, density functional theory (DFT) calculations, and ensemble-based energy scoring.
- To assess the utility of deep learning models, such as AlphaFold3, in analyzing enzyme variants and predicting activity.
Main Methods:
- Generation of single-point mutations in Pseudomonas aeruginosa LipA based on active site proximity.
- Evaluation of mutant variants using a computational pipeline including MD simulations, DFT calculations, and energy scoring.
- Application of deep learning models (AlphaFold3) for post hoc structural analysis.
- Reaction pathway analysis to determine rate-limiting steps and energy barriers.
Main Results:
- Several active enzyme variants were identified, but ranking them by activity using structural features proved challenging.
- Deep learning models produced highly similar active site geometries across variants, failing to differentiate activity levels.
- Reaction pathway analysis showed small energy barrier variations (5-15 kcal/mol) that were indistinguishable by current computational methods.
Conclusions:
- Current computational and deep learning methods have limitations in resolving subtle functional differences critical for incremental enzyme activity improvements.
- There is a need for improved benchmarks, reactive force fields, and more sensitive ranking metrics in computational enzyme design.
- Advancements are essential for designing enzymes with gradual, evolution-like activity enhancements and bridging the gap between structural prediction and catalytic function.
