Author ORCID Identifier
https://orcid.org/0009-0004-3823-4766
Date of Award
Summer 7-31-2026
Document Type
Thesis (Ph.D.)
Department or Program
Chemistry
First Advisor
Paul Robustelli
Abstract
Intrinsically disordered proteins (IDPs) lack a well-defined tertiary structure and instead populate heterogeneous conformational ensembles of rapidly interconverting structures. The populations of these states, their interconversion kinetics, and their interactions with molecular partners underlie the mechanics of biological functions, including molecular recognition, conformational transformations, and the formation of biomolecular condensates. In addition to function, IDPs are also implicated in the pathogenesis of many cancers and neurodegenerative diseases, often through aberrant aggregation. The dynamic behavior of IDPs and their interactions makes their characterization through classical experimental and computational methods notoriously difficult. IDPs occupy a myriad of diverse structures, leading to highly convoluted experimental signals and broadly distributed conformational spaces that are difficult to interpret and sample exhaustively with traditional simulation techniques.
This thesis develops and applies computational methods to characterize the structural and kinetic ensembles of IDPs by combining molecular dynamics simulations, statistical learning, experimental data, and artificial intelligence to obtain mechanistic descriptions of disordered protein behavior at atomic resolution. Deep learning-based kinetic models are used to characterize folding-upon-binding pathways in a fuzzy protein complex, resolving intermediate states and kinetic barriers between metastable bound conformations. Knot-theoretic, multiscale descriptors are developed to identify slow motions of IDPs, construct Markov state models of conformational dynamics, and build geometric deep learning methods for generative modeling of conformational ensembles. Integrative modeling approaches based on maximum-entropy reweighting are used to combine simulations with NMR and complementary experimental data, thereby enabling accurate atomic-resolution conformational ensembles. Drug discovery pipelines are designed to rapidly screen the binding affinities of small-molecule ligands to emergent IDP conformational states. Lastly, these approaches are combined to elucidate the mechanism by which small molecules stabilize oligomers of a phase-separating disordered protein region in the androgen receptor transactivation unit.
Taken together, this work advances a computational framework for studying proteins whose functions arise from heterogeneous conformational landscapes rather than a single dominant structure. By combining molecular simulation, experimental restraints, and machine learning, this thesis provides methods for resolving how disordered proteins populate, interconvert between, and interact through structurally diverse conformational ensembles.
Recommended Citation
Sisk, Thomas Robert, "CHARACTERIZING THE KINETICS AND THERMODYNAMICS OF INTRINSICALLY DISORDERED PROTEINS: DATA, GEOMETRY & MACHINE LEARNING" (2026). Dartmouth College Ph.D Dissertations. 533.
https://digitalcommons.dartmouth.edu/dissertations/533
Included in
Artificial Intelligence and Robotics Commons, Biological and Chemical Physics Commons, Data Science Commons, Dynamic Systems Commons, Non-linear Dynamics Commons, Numerical Analysis and Computation Commons, Numerical Analysis and Scientific Computing Commons, Statistical, Nonlinear, and Soft Matter Physics Commons
