Machine learning enhanced computational reverse-engineering analysis for scattering experiments (CREASE) for analyzing fibrillar structures in polymer solutions

In this talk, we present a machine learning (ML) enhanced computational reverse-engineering analysis of scattering experiments (CREASE) to analyze the small-angle scattering profiles of fibrillar structures with dispersity in diameter (e.g., aqueous solutions of methylcellulose (MC) fibrils). We first validate CREASE by taking as input scattering profiles of in-silico structures with known dimensions (diameter, Kuhn length) and outputting those known dimensions within error. We then show the improvement in speed of CREASE workflow through incorporation of artificial neural network (NN). Finally, we compare fibril diameter distributions determined by NN-enhanced CREASE from experimental scattering profiles of MC fibrils to those reported by Lodge, Bates, and coworkers (Macromolecules 2018) and fitted from analytical models. The similarity between the fibril diameter distributions determined by CREASE and analytical fitting corroborates the previous assertion that MC forms fibrils with consistent average diameters of ~15-20 nm. Our success in applying CREASE to experimental scattering profiles of complex structure and dispersity in dimension paves the way for its future application towards other unconventional fibrillar systems without appropriate analytical models.