Attana, Fedaa; (2025) Dynamic structural clustering of Class D β-Lactamases using Molecular simulations and Deep Learning. Doctoral thesis (Ph.D), UCL (University College London).

Text FA_THESIS_NO_SIGNATURES.pdf - Accepted Version Access restricted to UCL open access staff until 1 January 2027. Download (16MB)

Abstract

Class D β-lactamases (DBLs) have gained clinical significance as major contributors to antimicrobial resistance, particularly through carbapenem hydrolysis. While their structural flexibility is increasingly recognized as central to their activity, existing studies of DBL dynamics are fragmented: they focus on individual enzymes, employ inconsistent residue numbering, and thus remain difficult to compare. This thesis addresses these challenges in three stages. First, a literature consistent annotation framework is introduced. Using OXA-48 as a reference, the SAND (Structural Alignment-based Numbering of DBLs) scheme is developed together with a consensus secondary structure annotation, enabling homologous residues and elements to be consistently identified across the family. This framework resolves inconsistencies in the literature, supports reproducible analyses, and provides a resource for both experimental researchers and AI-based text/data mining tools. Second, a comparative dynamics study is conducted with the aim of filling the gap of dynamic knowledge among different OXAs and identifying measurable dynamical properties that can be linked to substrate profiles and functional phenotypes of DBLs. Enhanced sampling through adaptive bandit molecular dynamics simulations is used to explore conformational landscapes across representative enzymes, revealing conserved motifs and subfamily-specific differences in loop flexibility and hydrophobic bridge formation. Finally, deep learning is applied to analyze these large-scale simulations. Convolutional variational autoencoders were trained on inter-residue distance matrices, and the resulting latent representations were projected into low-dimensional spaces. This approach enabled the detection and interpretation of clustering patterns corresponding to metastable conformational states, which were mapped back to structural and functional determinants of substrate specificity. By integrating structural standardization, comparative dynamics, and deep learning–based interpretation, this thesis establishes a coherent framework linking sequence, structure, dynamics, and function in DBLs. The results consolidate fragmented knowledge, provide mechanistic insights into functional variability, and offer a foundation for rational inhibitor design.

Download activity - last month

Export as CSVJSONXML

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item