Dynamic and Thermodynamic Characterization of RNA-Protein Binding and Solvation Through Atomistic Simulations and Machine Learning Based Methods

Chakraborty, Amrita (2026) Dynamic and Thermodynamic Characterization of RNA-Protein Binding and Solvation Through Atomistic Simulations and Machine Learning Based Methods. PhD thesis, Indian Institute of Science Education and Research Kolkata.

Text (PhD thesis of Amrita Chakraborty (20RS045)) 20RS045.pdf - Submitted Version Restricted to Repository staff only Download (18MB)

Abstract

The intricate collaboration between RNA and proteins orchestrates a multitude of cellular processes, including transcription, translation, and intermediate stages of gene regulation across all species. Beyond these fundamental roles, various stages of the retroviral life cycle within the host are mediated through a spectrum of physical interactions—electrostatic, hydrogen bonding, van der Waals, hydrophobic, and stacking—between these two biomolecules. Understanding the microscopic mechanism underlying protein binding to RNA, and the influence of its functional environment comprising water and ions on the binding kinetics and thermodynamics, is therefore indispensable for therapeutic intervention in the replication of retroviruses such as immunodeficiency viruses. This thesis, using all-atomistic molecular dynamics simulations, free-energy calculations, and machine learning (ML) based methods, elucidates an integrated investigation into the molecular determinants of RNA–protein complex formation. Chapter 1 presents a structural and functional overview of RNA and proteins, emphasizing their dynamic association in biological regulation and highlighting the necessity of atomistic simulations to capture various functional dynamic states beyond the reach of static experimental techniques. Chapter 2 outlines the theoretical background and computational methodologies, describing MD protocols, free-energy analyses, dimensional-reduction schemes, and ML frameworks used throughout the work. The study conceptually delineates RNA–protein association into two interrelated phenomena—recognition, a kinetically controlled process, and binding, governed by thermodynamic stabilization. Chapter 3 first investigates the molecular mechanism of recognition using the Bovine Immunodeficiency Virus (BIV) TAR–Tat complex as a model. Simulations reveal intrinsic anticorrelated fluctuations between the loop and bulge domains of TAR RNA that are selectively exploited during Tat approach, enabling fluctuation-driven sensing and hierarchical domain communication. Chapter 4 focuses on the thermodynamic determinants of binding by comparing non-pathogenic BIV and pathogenic HIV TAR–Tat complexes. Despite structural similarity, the two exhibit distinct driving forces: BIV binding is enthalpy-dominated, whereas HIV-2 binding is entropy-favoured, consistent with its requirement of the host cofactor CycT1 for transcriptional activation. Building on these insights, Chapter 5 identifies four discriminating structural-dynamic descriptors—TAR integrity, loop–bulge distance, inter-domain correlation, and bulge length. A single bulge-nucleotide mutation (C24 insertion) distinguishing HIV-1 from HIV-2 reconfigures the RNA allosteric network, converting the mechanism from conformational selection to induced fit and enhancing Tat affinity. Chapter 6 extends the investigation to solvation dynamics, demonstrating that hydration and ionic reorganization actively modulate RNA structural fluctuations such as base flipping and thereby its protein binding behaviour. Decomposition of solvation-energy time-correlation functions reveals the intricate coupling between RNA motion and solvent–ion response, highlighting the multidimensionality of the process. Consequently, Chapter 7 develops an ML-based framework to identify optimal collective variables integrating both structural and environmental contributions, enabling construction of free-energy surfaces with accurate thermodynamic and kinetic representation of that functional base flipping event. Finally, Chapter 8 summarizes the overall findings and proposes the extension of this MD-integrated ML methodology toward building fully multidimensional free-energy landscapes of macromolecular recognition of RNA-protein, including explicit quantification of solvation free-energy contributions. Collectively, this thesis establishes a robust computational framework integrating molecular dynamics, statistical mechanics, and machine learning to characterize the thermodynamics and dynamics of RNA–protein association and solvation which can guide experimental strategies such as SELEX based methods to advance the rational design of antiviral therapeutics targeting RNA–protein recognition machinery.

Actions (login required)

View Item