(31i) Coupling Enhanced Sampling with Monte Carlo Techniques Improves Flexible-Backbone Protein Docking | AIChE

(31i) Coupling Enhanced Sampling with Monte Carlo Techniques Improves Flexible-Backbone Protein Docking


Harmalkar, A. - Presenter, Johns Hopkins University
Gray, J. J., Johns Hopkins University
Protein-protein interactions (PPIs) govern nearly all biological mechanisms, ranging from enzyme catalysis and inhibition to signaling and gene regulation. Understanding protein function and the associated dynamics of protein interactions at the molecular scale is key in delineating disease mechanisms, including COVID-19, cystic fibrosis, Alzheimer’s and cancer. Current computational tools are often confounded by the many degrees of freedom for structural flexibility of protein backbones, thereby hindering the accuracy of protein complex structure predictions. To overcome this challenge and to mimic binding-induced conformational changes in protein complexes, we integrated enhanced sampling methods, including temperature and Hamiltonian replica exchange, with docking protocols in Rosetta to improve local and global docking predictions. Our method allows us to rapidly explore relative orientations between protein partners with enhanced sampling. To capture backbone degrees of freedom, we sample backbone motions on residues comprising a putative interface on-the-fly, thereby incorporating binding-induced changes with docking. We benchmarked our protocol on 20 protein complexes starting from unbound binding partners. In a global search, i.e. without any prior knowledge of the binding site, we are able to reach lower energies and improve native structure discrimination compared to RosettaDock4 and ReplicaDock. With 2 x 106 MC steps employed for each target, we were able to identify near-native geometries using 300-500 CPU-hours, depending on the protein sizes, which is relatively efficient compared to conventional molecular dynamics. ­­In local coarse-grained docking, our method obtains sub-angstrom predictions for about half of the benchmark targets. This approach successfully samples global docking energy landscapes in rigid-body coordinates, and it captures several Angstroms of induced backbone conformational changes at the protein-protein interface.