Mots-Clés
cell simulations
Monte Carlo
Brownian dynamics
code development
C++
parallel computing
algorithms
Description
Recent developments in the field of protein structure prediction showed that protein models can routinely reach unprecedented levels of near-experimental accuracy. In this context, modeling protein interactions in the living cell is becoming more central than ever before. Despite impressive results in modeling protein complexes (cf CASP15 experiment) and alternative protein conformations, reliable sampling of transient (weak) protein-protein interactions and estimating the shape of the protein energy landscape is still out of reach for the general-purpose deep-learning architectures.
More classical techniques for modeling protein interactions include molecular docking and biomolecular simulations. While the latter can give access to the dynamics and the kinetics of the interactions, they are either relatively slow, if carried out at the all-atom representation, or largely coarse-grained, with one particle representing a protein. Consequently, there are only a few examples of simulations at the scale of the entire cell. Molecular docking methods are more efficient, especially those relying on systematic Fast Fourier Transform (FFT) sampling algorithms. However, they lack a reliable account of the kinetics of the association, they oversimplify solvation effects, and modeling the competition between several molecules is difficult in this framework. Due to these current limits in temporal and spatial resolutions, there has been a distinct lack of investigation on how the crowded environment of the cell impacts the physiological function of protein interactions in vivo.
Grounded on our preliminary results published in PNAS last year (Vakser, I. A.; Grudinin, S.; Jenkins, N. W.; Kundrotas, P. J.; Deeds, E. J. 2022), this project aims to address this gap through the development of a novel framework for modeling the dynamics of protein interactions in crowded environments combined with detailed experimental tests. We aim to bridge the two simulation approaches and reach unprecedented simulation timescales of milliseconds to seconds at all-atom resolution. To accomplish that, we will develop and apply Monte Carlo (MC) and Brownian Dynamics (BD) simulations to protein molecules in the all-atom representation. We will accelerate the computation of the interatomic potential using the FFT, assuming that some parts of the system can be approximated as rigid bodies, and thus their interactions can be pre-computed by systematic docking.
The candidate will
1) Generate protein-protein energy landscape in the cytosol. The candidate will develop novel methods for accurate docking of the cellular compounds to generate the intermolecular energy landscape. New diverse datasets of protein complexes will be generated for the development and benchmarking of docking. A systematic analysis of the intermolecular landscapes will provide a key framework for the predictive docking protocol. Global scan, scoring, and flexible refinement of the FFT docking will be advanced for adequate representation of the full intermolecular energy landscape, including the multiplicity of transient interactions.
2) Develop simulators of protein interactions in the cytosol. The candidate will develop novel MC and BD simulators of crowded environments to sample the energy landscape. The protocols will be further extended to proteins with conformational flexibility. Replica-exchange and parallelized versions of MC and BD will be developed. Tools for visualization and analysis of the simulation results will be developed.
3) Test the developed techniques by performing structural modeling of proteasome assembly kinetics in a dilute environment. The candidate will further proceed to model structural flexibility in the assembly, and the assembly in crowded environments. The results of modeling will be tested by experiment by our collaborators at UCLA (Eric Deeds lab).
We are looking for creative, passionate and hard-working individuals from applied math / computer science background with exceptional talent for computer science and mathematics and interest in computational physics and biology. Excellent oral, written and interpersonal communication skills are essential (working language will be English – knowledge of French is a plus). Excellent knowledge of computational physics and C++ is required. Knowledge of parallel programming / signal processing / machine learning / structural biology will be an asset.