Integrative computational modeling of protein interactions

Abstract

Protein interactions define the homeostatic state of the cell. Our ability to understand these interactions and their role in both health and disease is tied to our knowledge of the 3D atomic structure of the interacting partners and their complexes. Despite advances in experimental method of structure determination, the majority of known protein interactions are still missing an atomic structure. High‐resolution methods such as X‐ray crystallography and NMR spectroscopy struggle with the high‐throughput demand, while low‐resolution techniques such as cryo‐electron microscopy or small‐angle X‐ray scattering provide data that are too coarse. Computational structure prediction of protein complexes, or docking, was first developed to complement experimental research and has since blossomed into an independent and lively field of research. Its most successful products are hybrid approaches that combine powerful algorithms with experimental data from various sources to generate high‐resolution models of protein complexes. This minireview introduces the concept of docking and docking with the help of experimental data, compares and contrasts the available integrative docking methods, and provides a guide for the experimental researcher for what types of data and which particular software can be used to model a protein complex.

Combination of Ambiguous and Unambiguous Data in the Restraint-driven Docking of Flexible Peptides with HADDOCK: The Binding of the Spider Toxin PcTx1 to the Acid Sensing Ion Channel (ASIC) 1a

“…Potential sources of data for restraints include ASM, NMR titrations, crosslinking experiments, and bioinformatics predictions. 13,[17][18][19][20][21][22] Of these methods, data from ASM of the peptide is perhaps the most readily obtainable. 23,24 ASM can be used to identify residues that are critical for binding and thus likely to form part of the complex interface.…”

Structural prediction of protein models using distance restraints derived from cross-linking mass spectrometry data

“…Third, the XL-derived distance restraints are applied during the docking of the subunits into the model of the intact complex. The programs I-TASSER (iterative threading assembly refinement)9,10 and HADDOCK (high-ambiguity-driven protein–protein docking)11–13 were selected for protein structural prediction and docking, respectively, because (i) they both allow distance restraints to be specified by the user, and (ii) they were identified as top-scoring modeling algorithms in molecular-modeling challenges14,15. Moreover, HADDOCK allows conformational change of the individual subunits during complex docking of the protein backbone as well as of the side chains, therefore accounting for some protein flexibility.…”

The importance of dynamics in integrative modeling of supramolecular assemblies

“…Initially designed to employ NMR data, it has also been used with evolutionary data predicting inter-protein contacts [8,68] and adapted to handle SAXS data [69]. Conformational flexibility is here accounted in two main ways; that is, by providing ensembles from NMR structures, collections of X-rays structures or simulation snapshots, rather than static structures to the search algorithm [3], and by performing multidomain docking [70]. This latter scheme, like early programs such as FlexDock [71] and RNAbuilder [44], splits a flexible binding partner into subdomains based on an elastic network model and treats the parts independently, but under the strong constraint that the two halves of a hinge must be spatially close in the complex [70].…”

“…Although recent advances in cryo-electron microscopy (cryo-EM) are enabling near-atomistic resolution of large assemblies [2], a broad array of experimental methods can provide lower resolution data about overall shape, symmetry, composition, contact sites between constituent molecules, angular and distance restraints between domains, as extensively reviewed in [3]. In this context, integrative modeling attempts to consistently combine these heterogeneous, and sometime incomplete, data with the structures of the individual subunits that constitute a complex in order to generate models at near atomistic resolution [4].…”