Abstract
SummarySmall-angle X-ray scattering (SAXS) method is widely used in investigating protein structures in solution, but high-quality 3D model reconstructions are challenging. We present a new algorithm based on a deep learning method for model reconstruction from SAXS data. An auto-encoder for protein 3D models was trained to compress 3D shape information into vectors of a 200-dimensional latent space, and the vectors are optimized using genetic algorithms to build 3D models that are consistent with the scattering data. The program has been tested with experimental SAXS data, demonstrating the capacity and robustness of accurate model reconstruction. Furthermore, the model size information can be optimized using this algorithm, enhancing the automation in model reconstruction directly from SAXS data. The program was implemented using Python with the TensorFlow framework, with source code and webserver available from http://liulab.csrc.ac.cn/decodeSAXS.
Highlights
Small angle X-ray scattering (SAXS) from protein molecules in solution is a powerful technique, providing information on molecular structures and dynamics (Putnam et al, 2007; Grant et al, 2011; Svergun & Koch, 2003)
Model reconstruction will be advanced if the following criteria are met: (1) diverse shapes of 3D models can be efficiently represented to cover a broader range than those in structure databases; and (2) SAXS profiles can be computed for each model that can be scaled to arbitrary sizes
We first demonstrate that the auto-encoder works nicely in representing the shape information in the compressed format
Summary
Small angle X-ray scattering (SAXS) from protein molecules in solution is a powerful technique, providing information on molecular structures and dynamics (Putnam et al, 2007; Grant et al, 2011; Svergun & Koch, 2003). Because the solution scattering method does not require special treatment for protein molecules, such as crystallization in diffraction measurement or isotope labelling in nuclear magnetic resonance, SAXS experiments can be performed in high-throughput manners (Hura et al, 2009). Another major advantage of SAXS experiments is the ability to probe the structure and dynamics in solution, especially when combined with pumping methods to promote conformational changes (Neutze & Moffat, 2012; Kim et al, 2012). A database of shapes abstracted from actual protein complexes and efficiently represented using 3D Zernike polynomials was used to quickly retrieve 3D models that match experimental SAXS profiles, as implemented in sastbx.shapeup (Liu, Hexemer et al, 2012). We provide a solution to achieve this using an auto-encoder method combined with 3D Zernike representations (Canterakis, 1999; Liu, Morris et al, 2012)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
