Combined Bayesian and Normal Mode Flexible Fitting with Hamiltonian Monte Carlo Sampling for Cryo Electron Microscopy

Abstract : Density volumes obtained by three-dimensional reconstruction of biomolecular complexes from cryogenic electron microscopy (cryo-EM) images (also known as cryo-EM maps) can be interpreted in terms of atomic positions by flexible fitting. The fitting modifies an available atomic structure to match the target EM map. The most accurate fitting methods are based on atomic-coordinate degrees of freedom (e.g. Bayesian flexible fitting) but come with high computational cost for large required displacements. To reduce the computational cost, methods based on Normal Modes Analysis (NMA) decrease the number of degrees of freedom to only several collective motions (described by normal modes). The NMA-based methods are well-suited for global atomic displacements (large collective motions) but are suboptimal regarding local atomic displacements. To take advantages of both methods, we propose to combine them. We tested our method using synthetic and experimental cryo-EM maps of a complex with large-scale conformational changes (p97 ATPase). We show that the combination of both approaches efficiently performs global and local atomic displacements and that it can be more efficient and precise than any of the two approaches alone. To the best of our knowledge, this is the first method combining Bayesian and normal mode flexible fitting approaches.