(181b) Coupling Replica Exchange with Backbone sampling Captures Conformational Changes on Protein-Protein Interfaces | AIChE

(181b) Coupling Replica Exchange with Backbone sampling Captures Conformational Changes on Protein-Protein Interfaces


Harmalkar, A. - Presenter, Johns Hopkins University
Gray, J. J., John Hopkins University
Protein-protein interactions (PPIs) are involved in almost all biological processes in human health and disease. Understanding the structure of a protein complex can reveal biological mechanisms and suggest intervention strategies. Since experimental techniques are often infeasible, docking provides an alternative to elucidate structures and guide molecular engineering. Recently, AlphaFold (Jumper et al. 2021; Evans et al. 2021) and other deep learning algorithms (Baek et al. 2021) have made a breakthrough in the field of protein sequence-to-structure prediction. Yet, modeling protein interfaces and predicting conformational changes, remains an unsolved challenge, especially when there is significant conformational change in one or both binding partners. To overcome these limitations, we introduce a new enhanced sampling algorithm, called Hamiltonian-Resolution replica exchange (H-ResEx). The H-ResEx algorithm improves canonical sampling over rugged, atomistic energy landscapes by swapping conformations between the coarse-grained (CG) and all-atom (AA) states. To capture large-scale conformational changes, we mimic induced-fit approach of protein binding and sample backbone moves on-the-fly. We demonstrate the performance of our method for docking protein targets on a benchmark set of 100 proteins with moderate to high flexibility (unbound to bound RMSD over 1.2 Å up to 8 Å). Moreover, our advanced simulation approach equips the computational advantages of coarse-graining and requires 300-500 CPU hours for a docking simulation (depending on protein sizes), which is efficient compared to molecular dynamics-based approaches. With this work, we show that a CG/AA exchange scheme with conformational sampling shows substantial promise towards quantitative and qualitative modeling of protein complexes and their dynamic interactions. The proposed algorithm paves the way towards challenging applications in enhanced sampling of biomolecular systems, employing coarse-grained models for capturing conformations without compromising on the accuracy of the all-atomistic models.