Development of likelihood-based methods in structural biology

A new set of statistical tools will help biologists solve the structures of proteins and other large molecules from fuzzy or incomplete data. The problem is that many important biological molecules are difficult to image. X-ray crystallography can fail when the data are weak or the starting model is poor, and cryo-electron microscopy (cryo-EM) often produces blurry reconstructions at low resolution. This project adapts powerful likelihood-based statistical methods—already successful in crystallography—to tackle these stubborn cases. If successful, the work will automate the clustering of alternative molecular models into realistic ensembles, improve the detection of heavy-atom substructures for phasing, and develop better ways to dock atomic models into low-resolution cryo-EM maps. It will also explore how errors propagate through cryo-EM reconstructions and test a new interactive molecular dynamics approach for modelling structures at low resolution. This is fundamental science. It does not aim at an immediate practical application. But the methods it produces will directly affect how structural biologists determine the shapes of molecules that underpin drug design, enzyme engineering, and our understanding of disease mechanisms. Past work on likelihood-based crystallography has already transformed the field; this extension to cryo-EM could do the same for a technique now central to structural biology.