Modeling of proteins with low-resolution experimental data

SNIC 2019/2-26


SNIC Large Compute

Principal Investigator:

Erik Lindahl


Stockholms universitet

Start Date:


End Date:


Primary Classification:

10603: Biophysics

Secondary Classification:

10602: Biochemistry and Molecular Biology

Tertiary Classification:

10407: Theoretical Chemistry



We seek to explain how complex biomolecules move between different conformations, steered e.g. by ligand binding, voltage gradients, active transport, and how this relates to biological function. We use both simplified models and molecular simulations in combination with experiments, in particular electrophysiology, cryo-EM, and neutron scattering. This makes it possible to study features not readily available in experiments, such as transitions between states and finding new ways to use low-resolution experimental data. We are particularly interested in channels such as the GABA receptors and prokaryotic relatives (GLIC), where our goal is to understand the transitions between open, closed, and desensitized states, how small drugs allosterically modulate this process, and understand features such as the “alcohol cutoff effect” observed experimentally. We believe the transitions happen asymmetrically, and want to use simulations to study how the free energy of transitions depend on whether subunits move synchronously or not - this will be based on transition pathways we recently obtained from Langevin dynamics of principal components of experimental structures, combined with Markov State Models to sample the transition landscape. Cryo-EM is becoming an important research direction for us, and we have spent significant efforts on developing new GPU-accelerated methods for density reconstruction from micrographs in the RELION code, in particular to determine higher-resolution models of GLIC in several open/closed states, and ensembles containing many states - which will also be important to help determine what happens in experiments during plunge-freezing (in particular whether there is significant quenching to low-energy structures), which is an open question for cryo-EM in general. We will also combine cryo-EM data with simulations, and use new methods to apply Bayesian restraints to proteins (e.g. ribosomes and ion channels) based on electron density. Many of these structures are not well represented with a single all-atom model in blurred regions of large conformational flexibility or low resolution, but our approach provides an ensemble of conformations most likely to have generated the experimental data, instead of a single model. We will also use simulations to compute small-angle neutron scattering (SANS) spectra from open/closed-state channels and compare with experimental SANS data we are collecting in Grenoble. While SANS itself is extremely low resolution (essentially just describing shape), it provides the first accurate data about channel structure at room temperature, and comparing to simulations of existing structures will enable us to determine how well these correspond to functional channels (a matter of long debate in the field). In a second step, we will also attempt to refine channels against SANS data. Finally, we are continuously developing GROMACS, with the 2019 beta just released; this has turned into one of the most widely used codes in the world for molecular simulation, and SNIC resources are critical to help us tune performance, parallelization and scaling for state-of-the-art machines, not to mention that we are developing a number of algorithms such as efficient lattice summation methods for Lennard-Jones interactions.