Subject Area
Biophysics, Chemistry
Abstract
Pathogen resistance to β-lactam antibiotics compromises effective treatments of superbug infections. One major source of β-lactam resistance is the bacterial production of β-lactamases, which could effectively hydrolyze β-lactam drugs. In this thesis, the hydrolysis of various β-lactam antibiotics by class A serine-based β-lactamases (ASβLs) were investigated using hybrid Quantum Mechanical / Molecular Mechanical (QM/MM) minimum energy pathway (MEP) calculations and explainable machine learning (ML) approaches. The TEM-1/benzylpenicillin acylation reaction with QM/MM chain-of-states reaction pathways was firstly revisited. I proposed two decomposition methods for energy contribution analysis based on perturbing ML regression models. Both methods were shown to be model implementation invariant and successfully bridged the discrepancies between two pioneering mechanistic studies. The Toho-1 ASβL acylations of ampicillin and cefalexin were then investigated. I reported that the acylation pathway selection can be ligand dependent: ampicillin could undergo acylation via Lys73 or Glu166 acting as the general base while cefalexin acylation is limited to Lys73 as the general base. An explainable artificial intelligence (XAI) method, the Boltzmann-weighted Cumulative Integrated Gradients (BCIG), was developed to explain the different acylation pathway viability found for ampicillin and cefalexin. Lastly, conformational factors determining the GES-5/imipenem deacylation activity was investigated using edge-conditioned convolutional graph-learning (GL) methods. Critical vi mechanistic insights were derived from perturbative response of the GL latent representations, which explained the different deacylation reactivity between the two imipenem pyrroline tautomer states and identified the orientation of the carbapenem 6α-hydroxyethyl as the key factor that impacts the deacylation barrier heights. In summary, my thesis focuses on bridging QM/MM chain-of-states reaction pathway calculations and explainable ML to derive essential mechanistic insights into β-lactam resistance driven by ASβLs.
Degree Date
Spring 2022
Document Type
Dissertation
Degree Name
Ph.D.
Department
Chemistry
Advisor
Peng Tao
Number of Pages
99
Format
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License
Recommended Citation
Song, Zilin, "Unraveling Molecular Mechanisms of Antibiotic Resistance Through Multiscale Simulations and Explainable Machine Learning" (2022). Chemistry Theses and Dissertations. 29.
https://scholar.smu.edu/hum_sci_chemistry_etds/29
