Machine Learning Classification Model for Functional Binding Modes of TEM-1 β-Lactamase

Abstract

TEM family of enzymes is one of the most commonly encountered β-lactamases groups with different catalytic capabilities against various antibiotics. Despite the studies investigating the catalytic mechanism of TEM β-lactamases, the binding modes of these enzymes against ligands in different functional catalytic states have been largely overlooked. But the binding modes may play a critical role in the function and even the evolution of these proteins. In this work, a newly developed machine learning analysis approach to the recognition of protein dynamics states was applied to compare the binding modes of TEM-1 β-lactamase with regard to penicillin in different catalytic states. While conventional analysis methods, including principal components analysis (PCA), could not differentiate TEM-1 in different binding modes, the application of a machine learning method led to excellent classification models differentiating these states. It was also revealed that both reactant/product states and apo/product states are more differentiable than the apo/reactant states. The feature importance generated by the training procedure of the machine learning model was utilized to evaluate the contribution from residues at active sites and in different secondary structures. Key active site residues, Ser70 and Ser130, play a critical role in differentiating reactant/product states, while other active site residues are more important for differentiating apo/product states. Overall, this study provides new insights into the different dynamical function states of TEM-1 and may open a new venue for β-lactamases functional and evolutional studies in general.

Dynamical Behavior of β-Lactamases and Penicillin- Binding Proteins in Different Functional States and Its Potential Role in Evolution

“…Ser70 is a critical residue for TEM-1 catalytic activity, and is located on the loop of residues 64 to 74. The Ser70 is covalently bound to the ring opening intermediate of Penicillin G as a product of the acylation step [ 4 , 26 ]. Interestingly, the loop of residues 64 to 74 display significant flexibility only in the reactant state of TEM-1 ( Figure 5 ), showing its unique role for catalysis in the reactant state.…”

“…In one of our previous studies, the dynamical properties of TEM-1 in different functional states, including complexes binding with penicillin G and its de-acylation product, respectively, and the apo state were characterized through MD simulations and machine learning methods [ 26 ]. The key residues for TEM-1 dynamics in different functional states were identified using machine learning methods.…”

Allosteric Regulation at the Crossroads of New Technologies: Multiscale Modeling, Networks, and Machine Learning

“…Functional residues in residue networks are often connected via strong evolutionary relationships (Lockless and Ranganathan, 1999 ; Suel et al, 2003 ; Halabi et al, 2009 ; Aguilar et al, 2012 ; McLaughlin et al, 2012 ; Simonetti et al, 2013 ). Coevolution of protein residues can reflect correlated functional dynamics of these sites in mediating residue-residue contacts (Socolich et al, 2005 ), protein folding transitions (Morcos et al, 2011 ), and allosteric signaling in protein complexes (Wang et al, 2019 ). Coevolving residues could also form direct communication paths in the interaction networks with connections weighted according to dynamic couplings and coevolutionary interaction strengths between nodes (Chakrabarti and Panchenko, 2009 , 2010 ; Nishi et al, 2011 ).…”

“…These residue response maps could accurately describe how different protein residues are affected by allosteric perturbations exerted on the protein system. Another ML-based analysis of protein dynamics was employed to compare the binding modes of TEM-1 β-lactamase in different catalytic states (Wang et al, 2019 ). While conventional analysis methods including principal components analysis (PCA) could not differentiate TEM-1 in different binding modes, neural network models resulted in an excellent classification scheme that differentiated between closely related functional states (Wang et al, 2019 ).…”

Unraveling the energetic significance of chemical events in enzyme catalysis via machine-learning based regression approach

“…Computational methods have been employed to further illustrate the detailed TEM-1 catalytic mechanism [16][17][18][19][20][21][22][23] . Hybrid quantum mechanics/molecular mechanics (QM/MM) and molecular dynamics (MD) studies have validated the acylation mechanism and provided reaction pathways on the potential energy surface (PES) [16][17][18] .…”