Cryo-electron microscopy (cryo-EM) has become a key technique for resolving structures of large biomolecules. With the growing availability of computational modeling tools like AlphaFold, more precise atomic models are now accessible. Consequently, fitting models into cryo-EM maps has become a standard method for structure interpretation. However, accurately fitting these models remains challenging, especially with intermediate or low-resolution maps and large molecules of over a few thousand amino acid residues. Here, we present DMcloud, which addresses these challenges by combining diffusion models with local point cloud-based matching. Unlike conventional global fitting, DMcloud is designed for situations where AlphaFold2 (AF2) models capture accurate local domains but contain misoriented or misplaced regions. By converting both the AF2 model and cryo-EM map into point clouds, DMcloud performs iterative local alignments and denoising to refine model placement. DMcloud builds on our earlier method, DiffModeler, which performs global fitting of protein complexes using diffusion-based backbone tracing and AlphaFold-guided model assembly. We evaluated DMcloud on 71 intermediate- and 50 high-resolution cryo-EM maps using AF2 models. Across benchmarks, DMcloud consistently outperformed traditional fitting methods, particularly in low-resolution cases and where AF2 models contained significant structural discrepancies. DMcloud provides an accurate, automated framework for cryo-EM model fitting and is particularly effective in challenging cases where predicted models contain inaccuracies, such as domain misorientations. By focusing on local structure alignment using point cloud representations, DMcloud achieves higher accuracy in model placement than conventional methods. Its ability to selectively identify and adjust only the well-supported regions of given structure models makes it useful for large assemblies and partially resolved complexes. The method is freely available through our web server at https://em.kiharalab.org.
Terashi et al. (Sun,) studied this question.