Integrating Arrhenius Constraints with Lineage-Aware Meta-Learning for Few-Shot Prediction of Temperature-Dependent...

Area of Science:

• Biocatalysis and Enzyme Engineering
• Computational Biology
• Protein Science

Background:

• Enzyme engineering requires understanding temperature-dependent activity (kcat) for balancing performance and stability.
• Predicting enzyme kinetics is difficult due to limited multi-temperature data and data-driven models violating thermodynamic principles.

Purpose of the Study:

• To present PIMetaKcat, a hybrid computational framework for accurate prediction of enzyme kinetic parameters.
• To address the limitations of existing models in predicting enzyme activity across various temperatures.

Main Methods:

• Developed PIMetaKcat, a framework integrating phylogenetic information with first-principles thermodynamic constraints.
• Utilized lineage-aware meta-learning to capture family-specific kinetic patterns for predicting distant homologues.
• Enforced Arrhenius equation consistency to reconstruct full activity profiles, including optimal temperature (Topt) and activation energy (Ea).

Main Results:

• PIMetaKcat demonstrated high fidelity (0.957 ± 0.021, 0.565 ± 0.032) on a low-redundancy benchmark.
• Achieved superior generalization to low-similarity sequences compared to existing baseline models.
• Successfully distinguished distinct thermal niches and enabled high-precision ranking of mutant libraries.

Conclusions:

• PIMetaKcat offers a physics-anchored approach to accelerate the Design-Build-Test cycle for rational enzyme design.
• The framework provides accurate predictions of temperature-dependent enzyme kinetics, enhancing biocatalyst engineering.
• Openly available code and data facilitate further research and application in enzyme discovery.