Postdoctoral position

Machine learning for cryogenic electron microscopy

Apply here:
Deadline: May 13, 2022

Project description

Single-particle cryogenic electron microscopy (cryo-EM) is a tomographic method for computationally recovering the 3D structure of a biomolecule from transmission electron microscope (TEM) imaging data, which is having profound impact on structural biology in general, and drug design in particular. Recently, for example, the structure of the SARS-CoV-2 spike protein was recovered using cryo-EM methods.

Despite having enjoyed significant success, current 3D reconstruction methods used for guiding the assembly of an atomic model (i.e., the locations of atoms in the molecule) only work for quite rigid biomolecules. However, biomolecules often have inherent flexibility that is of interest in structural biology as changes in the atomic structure are crucial in determining function. Applying standard reconstruction methods on cryo-EM data from such flexible molecules produces 3D maps of insufficient resolution for fitting an atomic model. This project will develop methods for recovering the dynamics of atom models for flexible biomolecules. The input consists of 2D TEM images from cryo-EM, and the amino acid sequence as known from sequencing.

The first aim of this project is to construct data-driven priors of atomic models trained using simulations obtained from molecular dynamics. These priors will be constructed using deep neural networks inspired by the recent success of AlphaFold2 and others. Given such priors, it will then be possible to estimate the posterior distribution of atomic model trajectories given cryo-EM data. We will then attempt to sample from the entire posterior distribution, providing a measure of uncertainty in the estimated motions. The result will be accurate and robust methods for characterizing the dynamics of biomolecules in cryo-EM data.

Research environment

The position will be physically located at the Department of Mathematics at KTH and will be working closely with colleagues at KTH but also at SciLifeLab. In particular, this post-doc will be twinned with another postdoctoral researcher at SciLifeLab under the supervision of Erik Lindahl. The project is thus situated at the interface of applied mathematics and computational cryo-EM.