Integrating Macromolecular X-ray Diffraction Data with Variational Inference

X-ray crystallography is a fundamental technique for determining the three- dimensional structure of biological macromolecules. The method involves crys- tallizing the molecule, exposing it to X-ray beams, and recording the resulting diffraction patterns, which consist of discrete signals called reflections. A critical computational step is the integration of diffraction data, where reflection intensities must be accurately estimated from noisy images.We propose a deep learning-based integration algorithm that leverages variational inference to infer the intensity and background of each reflection under a Poisson- likelihood model. Preliminary results show that our model’s integrated intensities underperform on crystallographic metrics compared to state-of-the-art algorithms. However, our integrated intensities were successfully used to reconstruct an electron density map and atomic model.