Computational Reverse-Engineering Analysis for Scattering Experiments (CREASE) on Biomacromolecular Assemblies

Many biomacromolecules are known to undergo thermoresponsive assembly into various morphologies, making them promising candidates for temperature-modulated applications such as drug delivery and tissue engineering. As two examples, methylcellulose (MC) chains self-assemble to form fibrillar structure in aqueous solutions at elevated temperature [Schmidt et al. Macromolecules, 2018, 51, 7767-7775]; elastin-like peptide-collagen-like pepetide (ELP-CLP) conjugates self-assemble to form vesicles at physiological temperatures [Luo and Kiick, J. Am. Chem. Soc. 2015, 137, 15362-15365]. Understanding the assembled macromolecular structure - domain shapes and dimensions - is key to precise control of the behaviors of such biomacromolecular materials in commercial applications. Small-angle scattering (SAS) is a useful experimental technique to characterize the structure of polymer assembly in solutions. However, interpretation of the SAS output - a scattering profile of averaged intensity (I(q)) at various wave factors (q) - often relies on analytical models only available for several conventional structures. Interpretation of I(q) vs. q plots for unconventional/non-equilibrium assembled structures with simple/complex polymer conformations remains challenging as relevant analytical models are not always available for such cases. In this talk, we present our recently developed computational method, CREASE, to analyze SAS results obtained from structures with high dispersity in dimensions formed upon assembly of complex polymer chemistries. Using genetic algorithm (GA) as the optimizer, CREASE takes as input I(q) vs. q plot and a general category of assembly morphology (e.g., “semiflexible fibril” or “bilayer vesicle”) and provides as output the relevant dimensions of the structure and dispersity in the dimensions whose computed scattering profile most closely matches input scattering profile. We test the performance of CREASE on both scattering profiles of methylcellulose fibrils and ELP-CLP vesicles and compare the predicted dimensions with those obtained from fitting analytical models. We also show how CREASE can be accelerated by incorporating neural network (NN) models. This talk will demonstrate CREASE’s potential to extract useful information about microscopic structure from SAS outputs without being limited by off-the-shelf scattering models and greatly improve the interpretability of small-angle scattering as a method to characterize polymer structures.