Crystal centering using deep learning in X-ray crystallography

A key challenge in X-ray crystallography is to find a good point on the crystal on which to center the beam because the crystal takes radiation damage after a number of shots which significantly distort the measurements. Therefore, the beam needs to be aimed manually by an operator, which results in significant additional effort and time. This paper presents an approach toward automating the beam aiming using machine learning, training a neural network with labeled data, resulting in a more efficient system that does not rely on manual supervision to determine where to aim the beam. A range of different neural network architectures are evaluated based on the accuracy of their predictions.