Although electron cryo-microscopy (cryo-EM) has recently achieved resolutions of better than 3 Å, at which point molecular modeling can be done directly from the density map, analysis and annotation of a cryo-EM density map still primarily rely on fitting atomic or homology models to the density map. In this article, we present, to our knowledge, a new method for flexible fitting of known or modeled protein structures into cryo-EM density maps. Unlike existing methods that are guided by local density gradients, our method is guided by correspondences between the α-helices in the density map and model, and does not require an initial rigid-body fitting step. Compared with current methods on both simulated and experimental density maps, our method not only achieves greater accuracy for proteins with large deformations but also runs as fast or faster than many of the other flexible fitting routines.

We present a method for classifying the shape of middle cerebral artery (MCA) aneurysms using segmented surfaces from angiograms. The classification follows a set of visual criteria established by experienced surgeons to group aneurysms based on the clipping strategies used in surgery. Starting from a centerline representation of the input, our method automatically classifies the input into one of 4 types using a combination of graph analysis and supervised learning. When evaluated on a cohort of 84 subjects, our method achieves between 60% to 69% expected classification accuracy (p < 10 −3 ) with zero to a moderate amount of input from novice users.