Two independent projects will be carried out under this allocation. Project 1 concerns high-resolution structural studies of amyloid and amyloid-like protein assemblies by cryo-EM. We will determine structures of phenol-soluble modulins (PSMs), with emphasis on PSMβ fibrils, and extend ongoing work on human islet amyloid polypeptide (hIAPP) polymorphism. hIAPP aggregation is linked to loss of insulin-producing β-cells in type 2 diabetes, and our previous results show that small sequence changes can produce distinct fibril architectures. The new work will investigate how chaperones, solvent electrostatics, and antibodies alter hIAPP fibril formation, polymorph selection, and surface recognition. Samples expected to contain multiple polymorphs will be processed by iterative 2D/3D classification, helical reconstruction, focused classification, and atomic model refinement. For hIAPP–antibody complexes, special emphasis will be placed on separating antibody-bound and unbound fibrils and on identifying conformational or occupancy heterogeneity at fibril surfaces. The structural outputs will provide molecular models of PSM assemblies, hIAPP polymorphs stabilized under different solution conditions, and antibody-recognition interfaces. Project 2 concerns computational chemistry of infrared probes, with focus on thiocyanate (SCN) labels. SCN labels are compact vibrational reporters whose nitrile-stretching frequencies respond to local polarity, hydrogen bonding, and electrostatic fields. However, measured frequency shifts and linewidths are not directly interpretable because they contain coupled contributions from electrostatics, hydrogen bonding, short-range exchange repulsion, dispersion, and probe dynamics. This project will develop predictive models that translate molecular dynamics trajectories into FTIR and 2D-IR spectra. First, AIMD and DFT single-point calculations will be used to generate reference energies, forces, dipoles, electrostatic potentials, and vibrational response quantities for SCN-containing systems in solvents and selected protein environments. These data will train DeepMD-type potentials and neural-network dipole/frequency models, enabling long trajectories with DFT-level accuracy and direct spectral simulations from dipole autocorrelations and local vibrational response functions. Second, an interpretable active-learning workflow will be developed for automatic construction of solvatochromic maps. Existing electrostatic charge-map models will be extended with radial-embedding descriptors that represent short-range contact geometry around the SCN probe, allowing the model to capture exchange-repulsion contributions that cannot be recovered from electrostatics alone. The models will be validated against experimental FTIR and 2D-IR spectra in simple solvents and then tested on SCN-labeled Ras/ralGDS systems at multiple labeling sites.