Identification of Domains in Protein Structures from the Analysis of Intramolecular Interactions

Abstract

The subdivision of protein structures into smaller and independent structural domains has a fundamental importance in understanding protein evolution and function and in the development of protein classification methods as well as in the interpretation of experimental data. Due to the rapid growth in the number of solved protein structures, the need for devising new accurate algorithmic methods has become more and more urgent. In this paper, we propose a new computational approach that is based on the concept of domain as a compact and independent folding unit and on the analysis of the residue-residue energy interactions obtainable through classical all-atom force field calculations. In particular, starting from the analysis of the nonbonded interaction energy matrix associated with a protein, our method filters out and selects only those specific subsets of interactions that define possible independent folding nuclei within a complex protein structure. This allows grouping different protein fragments into energy clusters that are found to correspond to structural domains. The strategy has been tested using proper benchmark data sets, and the results have shown that the new approach is fast and reliable in determining the number of domains in a totally ab initio manner and without making use of any training set or knowledge of the systems in exam. Moreover, our method, identifying the most relevant residues for the stabilization of each domain, may complement the results given by other classification techniques and may provide useful information to design and guide new experiments.

The value of protein structure classification information—Surveying the scientific literature

“…In cases where domain boundaries in CATH and SCOP differed, analysis was repeated with both definitions in order to ensure the robustness of their results: “The analysis was carried out using both CATH and SCOP in order to ensure that the ACO values that are calculated separately for each domain do not depend on the choice of domain boundaries that may differ in the two databases.” 13 studies used both the SCOP and CATH classifications for training or benchmarking algorithms (Use Category B); in 5 of these papers, this was done by using a consensus data set containing only data that was in agreement between the two databases, and in the rest, both databases were analyzed separately and results were presented for both. For example, a study by Genoni and colleagues that presented and evaluated an energy‐based method for identifying domains benchmarked the method on two datasets that consist of consensus domains from SCOP and CATH: the Benchmark_2 and Benchmark_3 datasets . The ThreaDom domain identification method developed by Xue and colleagues was mainly trained and evaluated on CATH domains, but also tested on SCOP domains to demonstrate “that the distinctive domain definitions of different databases have no impact on the training and testing procedures of domain predictions …”

The answer lies in the energy: how simple atomistic molecular dynamics simulations may hold the key to epitope prediction on the fully glycosylated SARS-CoV-2 spike protein

“…Otherwise put, putative interacting patches are hypothesized to be characterized by nonoptimized intramolecular interactions. Actual binding to an external partner such as an Ab is expected to occur if favorable intermolecular interactions determine a lower free energy for the bound than the unbound state 21,23,34 . Furthermore, minimal energetic coupling with the rest of the protein provides these subregions with greater conformational freedom to adapt to a binding partner and improves their ability to absorb mutations without affecting the protein's native organization and stability in a way that could be detrimental for the pathogen: all these properties are indeed hallmarks of Ab-binding epitopes.…”

“…The less restrictive 15% cutoff subdivides the full-length, fully folded S protein into potentially immunoreactive domains (see Figure 1B,C and Methods) 22, 25,34 . The goal is to uncover regions that may normally be hidden from recognition by Abs in the native protein structure, but that can be experimentally expressed as isolated domains.…”

“…In the case of multidomain proteins such as S, the first eigenvector is not sufficient, and more eigenvectors are needed to capture the essential interactions for folding/stability and binding. The interaction matrix %& is thus decomposed instead via the alternative approach developed by Genoni et al 34 . In this scenario, the aim is to select the smallest set of / eigenvectors that cover the largest part of residues (i.e., components) with the minimum redundancy under the assumption that: (a) for each domain there should exist only one associated eigenvector recapitulating its most significant interactions; (b) each "domain eigenvector" has a block structure whereby its significant components correspond to the residues belonging to the identified domain; (c) combination of all significant blocks covers all residues in the protein.…”

Differential Antibody Recognition by Novel SARS-CoV-2 and SARS-CoV Spike Protein Receptor Binding Domains: Mechanistic Insights and Implications for the Design of Diagnostics and Therapeutics

“…Selection is carried out on the basis of a threshold value (called softness), which defines the percentage of the set of putative interaction sites by including increasing residue-residue coupling values until the number of couplings that correspond to the lowest contactfiltered pairs under the threshold was reached. Detailed Prediction of protein interaction surfaces: MLCEMLCE is a technique based on the analysis of the interaction energies of all the amino acids in a protein3,[28][29][30][31] . In particular, it computes the non-bonded part of the potential (van der Waals, electrostatic interactions, solvent effects) via a MM/GBSA calculation, obtaining, for a protein composed by residues, a × symmetric interaction matrix %& .…”