Machine Learning-Based Computational Methods for Interpreting Small Angle Scattering Data from Soft Materials

The design and characterization of soft macromolecular materials depend critically on understanding their complex assembled structures at various length scales. Traditional methods for analyzing structural characterization data from techniques like Small-Angle Scattering (SAS) often rely on fitting to approximate analytical models, which are frequently inadequate for novel polymer chemistries or unconventional assembled morphologies. In this talk, I will present Computational Reverse Engineering Analysis for Scattering Experiments (CREASE), an open-source computational method developed by my group to overcome these limitations with traditional analytical model interpretations. CREASE utilizes a genetic algorithm coupled with machine learning (ML) enhancements to interpret 1D or 2D profiles from SAXS or SANS experiments. I will share the key steps in this ML-enhanced CREASE workflow and explain how it identifies multiple sets of structural features whose computed scattering profiles match the input SAS-measured profile. I will also include specific case studies to demonstrate the successful application of CREASE to diverse soft materials, including amphiphilic polymer solutions forming unconventional micelles, fibrillar structures, and nanoparticle mixtures. I will also highlight an example that demonstrates the application of CREASE-2D, which interprets the entire 2D scattering pattern, providing unprecedented insight into both isotropic and anisotropic structural arrangements and orientational order that are missed by traditional 1D azimuthal averaging. By delivering detailed real-space structural representations and enabling the characterization of features beyond the scope of existing analytical models, the CREASE methods serve as a powerful tool for establishing structure-property relationships and ultimately guiding the rational design of new soft materials.