It is often a challenge to atomically determine the structure of large macromolecular assemblies, even if successfully crystallized, due to their weak diffraction of X-rays. Refinement algorithms that work with low-resolution diffraction data are necessary for researchers to obtain a picture of the structure from limited experimental information. Relationship between the structure and function of proteins implies that a refinement approach delivering accurate structures could considerably facilitate further research on their function and other related applications such as drug design.

Here a refinement algorithm called the Deformable Complex Network is presented. Computation results revealed that, significant improvement was observed over the conventional refinement and DEN refinement, across a wide range of test systems from the Protein Data Bank, indicated by multiple criteria, including the free R value, the Ramachandran Statistics, the GDT (<1Å) score, TM-score as well as associated electron density map.