Towards automated crystallographic structure refinement with a differentiable pipeline

The lack of interfaces between crystallographic data and machine learning methods prevents the application of modern deep learning frameworks to crystal structure determination. Here we present SFcalculator, a differentiable pipeline to generate crystallographic observables (structure factors) from atomistic molecular structures and a bulk solvent model. This calculator fills the gap between the long-established crystallography field and state-of-the-art deep learning algorithms. We discuss the correctness and performance of SFcalculator by comparing with the current most-used tool Phenix. Finally, we demonstrate with an initial try that it enables automated structure refinement in a well-regularized latent space defined by a deep generative model, providing a principled way to impose prior knowledge. We believe this tool paves the way towards fully automated structure refinement and a possible end-to-end model, which is crucial for the next generation of high-throughput diffraction experiments.