Accurate flexible refinement of atomic models against medium-resolution cryo-EM maps using damped dynamics 08 Information and Computing Sciences 0801 Artificial Intelligence and Image Processing

Abstract

Background: Dramatic progress has recently been made in cryo-electron microscopy technologies, which now make possible the reconstruction of a growing number of biomolecular structures to near-atomic resolution. However, the need persists for fitting and refinement approaches that address those cases that require modeling assistance. Methods: In this paper, we describe algorithms to optimize the performance of such medium-resolution refinement methods. These algorithms aim to automatically optimize the parameters that define the density shape of the flexibly fitted model, as well as the time-dependent damper cutoff distance. Atomic distance constraints can be prescribed for cases where extra containment of parts of the structure is helpful, such as in regions where the density map is poorly defined. Also, we propose a simple stopping criterion that estimates the probable onset of overfitting during the simulation. Results: The new set of algorithms produce more accurate fitting and refinement results, and yield a faster rate of convergence of the trajectory toward the fitted conformation. The latter is also more reliable due to the overfitting warning provided to the user. Conclusions: The algorithms described here were implemented in the new Damped-Dynamics Flexible Fitting simulation tool "DDforge" in the Situs package.