Bandy, LeahWojnar, AnnaPike, LoganHarrand, QuinnPankavich, StephenPak, Alexander J.2024-05-082024-05-08https://hdl.handle.net/11124/179043https://doi.org/10.25676/11124/179043Many groups of molecules exhibit self-assembly behavior to form large-scale hierarchical structures. Scientists are interested in identifying the molecular basis for self-assembly, but the spatial and temporal resolution of current experimental techniques precludes observation of assembly details at the nanoscale. Meanwhile, conventional all-atom computer simulations remain too costly, and coarse-grained simulations, which trade detailed information for lower computational complexity, remain difficult to apply to macromolecular assembly. We utilize a supervised dimension reduction approach called active subspaces to enable coarse-grained simulations of self-assembling systems at reduced computational cost (compared to all-atom) and increased accuracy (compared to other coarse-grained models). The active subspace method identifies the most important directions in an input parameter space that influence a corresponding output function. The goal of this project is to develop and formalize the active subspace framework to derive coarse-grained models from all-atom data for reversibly aggregating alanine peptides. Our strategy is to explore spherical harmonics as the input parameter space, corresponding to potential energies as the output. Preliminary results indicate that a reduced set of spherical harmonics can provide a descriptive basis useful for coarse-grained modeling and simulation.postersengCopyright of the original work is retained by the author.Text