Abstract
This thesis studies the complexities of protein dynamics and allostery, presenting methodologies that leverage combined computational approaches, including molecular modeling and machine learning, to investigate these phenomena.
Protein dynamics bridge conformational ensembles and their corresponding functional states. It further plays a critical role in a wide range of biological processes. Therefore, studying protein dynamics is essential for understanding how proteins fulfill their biological functions. Allostery plays a pivotal role in biological processes, acting as a fundamental mechanism through which proteins transmit signals and regulate their activity. Despite its critical importance in biology, the specific allosteric mechanisms governing most proteins remain elusive.
Employing a multifaceted computational approach, the first study initially delves into microsecond molecular dynamics (MD) simulations of AsLOV2, uncovering the critical role of β-sheets in mediating its conformational changes and allosteric signal transmission. By integrating simulation data with experimental findings and mutational analyses, key hydrogen bonds pivotal for allostery are identified.
Advancing the discourse, this thesis highlights the integration of machine learning (ML) and deep learning (DL) techniques for the study of protein allostery, including studies of allosteric mechanisms, prediction of allosteric effects and allostery-related physicochemical properties, and allosteric protein engineering. This leads to two pivotal projects that utilize ML and DL to further our understanding and capabilities in protein dynamics and allostery.
Automated Machine Learning (AutoML) is employed for the precise prediction of allosteric sites, showcasing a model that not only achieves a remarkable 82.7% ranking probability for identifying allosteric sites within the top three predictions, but also validates its robustness by making predictions for proteins beyond the initial dataset. The final model is deployed to the Protein Allosteric Sites Server to facilitate the research in this field.
The variational Autoencoder (VAE) model is rigorously evaluated for its ability to assist protein conformation exploration. It is demonstrated that VAE can retain critical properties in a high-dimensional conformational space and predict physically plausible conformations that are infrequently accessed through traditional MD simulations, thereby providing valuable seed conformations for initiating new simulation studies.
Further, MD simulations and a Markov state model are utilized to characterize the functional conformational states of these variants. This analysis focuses on changes in the receptor-binding domain of the SARS-CoV-2 spike protein, particularly the alterations in conformational mobility that might enhance the virus’s transmissibility and immune evasion. Integrated with further perturbation-based approaches, the analysis provides insights into potential immune escape mechanisms.
This work not only extends computational methodologies for probing protein allostery, but also presents adaptable workflows for addressing broader biochemistry challenges, marking a unique contribution to the computational study of protein dynamics and allostery.
Degree Date
Summer 2024
Document Type
Dissertation
Degree Name
Ph.D.
Department
Chemistry
Advisor
Peng Tao
Number of Pages
199
Format
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License
Recommended Citation
Xiao, Sian, "Computational Study of Protein Dynamics and Allostery through Molecular Modeling and Machine Learning" (2024). Chemistry Theses and Dissertations. 46.
https://scholar.smu.edu/hum_sci_chemistry_etds/46
