Machine learning enhanced CREASE method for Analyzing 2D Small Angle Scattering Profiles

Understanding structural diversity in polymeric materials is a key step towards engineering new materials for various applications. One way to probe structures at varying length scales (nm to micron) is through small angle scattering (SAS); where a typical measurement provides as output the scattered intensity (I) vs. the magnitude of the wavevector (q) and azimuthal angle (Ɵ). To overcome some of the challenges researchers have in interpreting this two-dimensional I(q, Ɵ), especially for structures exhibiting anisotropy, we present a machine learning boosted ‘computational reverse engineering analysis for scatting experiments’ (CREASE) method. The chosen machine learning model, XGBoost, is trained to relate structural features (e.g., particle shapes and sizes, orientational order in particle arrangement) to the I(q, Ɵ) profile. Using the trained XGBoost model within the CREASE workflow, we accelerate the identification of structural features whose computed I(q, Ɵ) matches the input (experimental) I(q, Ɵ). This streamlined XGBoost-CREASE methodology eliminates traditional complexities in manual interpretation and provides an efficient and fast way to understand structural diversity in soft materials.