Application of Bayesian statistics and machine learning to interpretation, refinement and validation of cryoEM maps as atomic models

Macromolecules play an important role in almost all the stages of all biological processes. Knowledge of their three-dimensional structures help understanding these processes, which in turn may help to fight diseases, e.g. by designing new drugs. X-ray crystallography, Nuclear Magnetic Resonance and now increasingly more and more single particle cryo electron microscopy techniques are the main methods for elucidation of 3D structures of biological macromolecules and molecular complexes. Focus of this project is on utilisation and extraction of maximum possible atomic information from the density maps derived using cryoEM techniques.
The purpose of this project is to develop and implement mathematical, statistical and computational techniques for the interpretation of cryoEM maps with atomic models. Computational techniques for the refinement and validation of the atomic models will also be developed and applied to real life cases. The main techniques acquired during and used for delivering the project aims are Bayesian statistics, optimisation of multidimensional functions and machine learning methods. In particular, machine learning techniques, such as deep learning, will be used to derive prior knowledge with the associated probability distributions from the atomic models and density maps available in the Protein and Electron Microscopy Data Banks. The derived prior knowledge will be applied for (semi)automatic interpretation of cryoEM maps with atomic models. They will also be used for refinement and validation of atomic models.
Developed techniques will be applied to the wide range of problems of structural biology in the MRC-LMB. The results will be presented in the international conferences and teaching workshops.