Biophysical Mechanisms and Computational Design Strategies for Modulating Amyloid Formation and Protein–Protein Interactions

Abstract: Amyloid fibrils are highly ordered protein assemblies with a characteristic cross–β architecture and are central to numerous human diseases, including Alzheimer’s disease, Parkinson’s disease, and other neurodegenerative and systemic disorders. Although amyloid formation has been studied extensively, effective strategies to control aggregation remain limited, partly due to an incomplete understanding of the molecular forces governing self-assembly and the lack of tractable, disease-relevant systems for therapeutic design. I investigate the biophysical mechanisms underlying amyloid formation and develop computational strategies to modulate aggregation pathways and protein–protein interactions.

At the molecular level, amyloid assembly is driven by intermolecular hydrogen bonding, side-chain packing into steric zippers, and the dewetting of protein interfaces. Using tau as a model system, I examine how mutations, isoforms, and local sequence context influence these interactions. Atomistic molecular simulations reveal that subtle sequence changes can shift the balance between intramolecular stabilization and intermolecular amyloid formation by altering hydrogen-bonding networks and local water structure, providing a mechanistic basis for differences in aggregation propensity.

Motivated by the central role of water in aggregation, I introduce a context-aware framework for modeling protein hydrophobicity based on dewetting free energies. This approach captures both enthalpic and entropic contributions to water removal and reveals a strong dependence on chemical and geometric context that is absent in traditional hydrophobicity scales. I develop rapid dewetting free energy predictors that enable efficient identification of hydrophobic interfaces relevant to amyloid growth and protein binding.

These biophysical insights are translated into computational design strategies for controlling aggregation. Protein structure prediction, diffusion-based design, and protein language models are used to create synthetic, seeding-competent mini-amyloids that reproduce disease-relevant interfaces and serve as experimentally tractable surrogates for patient-derived fibrils. These systems enable the computational design and prioritization of amyloid binders aimed at inhibiting aggregation.

Finally, protein language models are applied at the proteome scale to identify aggregation- and phase-separation-prone regions across intrinsically disordered proteins, revealing new patterns of self-assembly linked to cellular function and disease. Together, this work integrates biophysical modeling, machine learning, and protein design to advance understanding of amyloid formation and provide new strategies for modulating pathological protein aggregation.