Development of likelihood-based methods in structural biology (360G-Wellcome-209407_Z_17_Z)

I propose to build on our development of new likelihood-based methods for structural biology, transposing approaches that have had great impact in macromolecular crystallography to the new area of cryo-EM. 1) In crystallography we will enable the solution of difficult structures (poor data or poor starting models) that still evade current methods. New statistical innovations will automate the clustering of alternative molecular replacement models into sensible ensembles representing different conformations. 2) We will exploit a promising new approach to the determination of substructures for SAD phasing, based on our SAD likelihood function. 3) In cryo-EM we will investigate the propagation of errors in reconstructions, building on this understanding to devise improved likelihood-based methods to dock atomic models into cryo-EM maps, particularly those challenging cases determined at low resolution such as sub-tomogram averages. The implications of multi-variate cryo-EM likelihood targets will be explored, with potential applications in the angular deconvolution of cryo-EM maps. 4) Finally, we will develop a new approach to modelling macromolecular structures at low resolution, using interactive molecular dynamics flexible fitting to combine high-quality potential functions with likelihood targets.