Artificial neural networks for solution scattering data analysis

Abstract

Small-angle X-ray scattering (SAXS) experiments are widely used for the characterization of biological macromolecules in solution. SAXS patterns contain information on the size and shape of dissolved particles in nanometer resolution. Here we propose a method for primary SAXS data analysis based on the application of artificial neural networks (NNs). Trained on synthetic SAXS data, the feedforward NNs are able to reliably predict molecular weight and maximum intraparticle distance (D_max) directly from the experimental data. The method is applicable to data from monodisperse solutions of folded proteins, intrinsically disordered proteins, and nucleic acids. Extensive tests on synthetic SAXS data generated in various angular ranges with varying levels of noise demonstrated a higher accuracy and better robustness of the NN approach compared to the existing methods.

Keywords: SAXS; artificial intelligence; machine learning; neural networks.