Native or Non-Native Protein–Protein Docking Models? Molecular Dynamics to the Rescue

Abstract

Molecular docking excels at creating a plethora of potential models of protein–protein complexes. To correctly distinguish the favorable, native-like models from the remaining ones remains, however, a challenge. We assessed here if a protocol based on molecular dynamics (MD) simulations would allow distinguishing native from non-native models to complement scoring functions used in docking. To this end, the first models for 25 protein–protein complexes were generated using HADDOCK. Next, MD simulations complemented with machine learning were used to discriminate between native and non-native complexes based on a combination of metrics reporting on the stability of the initial models. Native models showed higher stability in almost all measured properties, including the key ones used for scoring in the Critical Assessment of PRedicted Interaction (CAPRI) competition, namely the positional root mean square deviations and fraction of native contacts from the initial docked model. A random forest classifier was trained, reaching a 0.85 accuracy in correctly distinguishing native from non-native complexes. Reasonably modest simulation lengths of the order of 50–100 ns are sufficient to reach this accuracy, which makes this approach applicable in practice.

Structural insights into recognition of chemokine receptors by Staphylococcus aureus leukotoxins

“…A summary of the information extracted from MD simulations is shown in Table 2 . For both complexes, we monitored the number of inter-chain hydrogen bonds (H-bonds), buried surface area (BSA) and iRMSD relative to the starting model, as these parameters were recently shown to be useful in discriminating native from non-native docked models by MD ( Jandova et al, 2021 ). We additionally outputted the time evolution of distances involving receptor sulfotyrosines and LukE site 1 and 2 residues ( Figure 6—figure supplement 1 and Figure 7—figure supplement 1 ), and visually inspected models for qualitative agreement with available mutagenesis data.…”

Computational epitope binning reveals functional equivalence of sequence-divergent paratopes

“…The Epibin score predicts competition for epitope binding based on pairwise similarities of complex models produced by docking. While many of the docking models are likely to be incorrect (with perhaps one near-native, some relatively close, and more well off-target [40] , [41] , [42] ), intuitively they rely on and thus reveal complementarity between a putative paratope and putative epitope. Therefore, if the docking models of two protein binders involve many of the same regions on the target, then the binders are likely to have similar binding preferences and compete for the (unknown) true epitope.…”

A Computational Approach to Investigate TDP-43 RNA-Recognition Motif 2 C-Terminal Fragments Aggregation in Amyotrophic Lateral Sclerosis

“…To characterize each proposed molecular complex, composed of two different fragment configurations, we performed statistical analysis on the dynamical behavior of each system, since it is known that good models have usually higher stability during MD simulations [ 27 , 28 ]. To this end, we performed 20 ns-long MD simulations [ 27 ] and analyzed the RMSD at equilibrium and the percentage of C α binding atoms in the docking complex that are preserved at the end of the simulation (as a function of increasing cut-off distance).…”

“…Indeed, MD simulations are a common tool to improve the quality of docking complexes by refining them [ 30 , 31 , 32 ], since they can account for conformational changes needed for binding at different levels, particularly on the scale of atoms, sidechains, loops, small molecules, or interfaces [ 27 ]. Thus, even complexes that lose part of initial (docking pose) contacts may assume a stable conformation during the dynamics refinement.…”