BPS2026 – Network Hamiltonian models for transient structure in intrinsically disordered proteins

Network Hamiltonian models (NHMs) represent molecular systems by topological coarse-graining, in which system elements are represented as nodes in a graph whose edges represent bound interactions. An effective Hamiltonian on the topological degrees of freedom is then used to model patterns of interactions, with terms representing multi-body forces; this can allow for computationally scalable and parsimonious modeling of large and complex systems. Successfully used to model phenomena such as protein aggregation and structural variation in crystal structures of globular proteins, NHMs have been suggested as useful tools for modeling transient structure in intrinsically disordered proteins (IDPs). In particular, network representations provide a flexible tool for capturing the conformational states of IDPs, without relying on the presence of conventional secondary structure. Here, we use NHMs on residue-level network representations to model conformational ensembles of ABeta [1-40], comparing wild-type and E22G variant trajectories obtained from atomistic molecular dynamics at multiple temperatures. We examine alternative corrections for the entropic effects of backbone constraints, as well as novel NHM terms to capture transient sheet-like and helix-like behavior. We also consider the use of infinite hierarchical mixture models (i.e., Dirichlet process mixtures) to represent IDP ensembles in terms of mixtures of conformational ensembles approximated by simpler, local potentials.