Deep learning insights into β-lactamase dynamics and resistance evolution.

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• Antimicrobial resistance driven by beta-lactamase enzymes is a growing global health threat.
• Understanding enzyme adaptation requires analyzing conformational dynamics, which traditional methods struggle to capture.

Purpose of the Study:

• To review advances in applying deep learning to study beta-lactamase conformational dynamics.
• To highlight how deep learning methods can reveal mechanisms of enzyme adaptation and resistance.

Main Methods:

• Convolutional variational autoencoders (CVAEs) for reconstructing conformational landscapes from simulations.
• DiffNets for identifying structural determinants of biochemical phenotypes.
• Geometric deep learning and graph neural networks for analyzing active-site plasticity and allosteric couplings.

Main Results:

• Deep learning exposes metastable states, cryptic pockets, and catalytic intermediates in beta-lactamases.
• Methods accurately predict mutational fitness and epistasis by capturing protein dynamics.
• Identified structural determinants of substrate specificity and inhibitor susceptibility.

Conclusions:

• Deep learning provides a powerful, unified framework for analyzing protein dynamics in enzyme adaptation.
• This approach offers predictive insights into beta-lactamase evolution and resistance mechanisms.
• Enables mechanistic understanding crucial for combating antimicrobial resistance.