Model reconstruction from Small angle X-ray Scattering data using deep learning methods

Abstract

We present an algorithm based on a deep learning method for model reconstruction from small angle X-ray scattering (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 algorithm was implemented using Python with the TensorFlow framework and tested with experimental data, demonstrating capacity and robustness of accurate model reconstruction even without using prior model size information.