Prediction of interface residues in protein-protein complexes by a consensus neural network method: test against NMR data

The number of structures of protein-protein complexes deposited to the Protein Data Bank is growing rapidly. These structures embed important information for predicting structures of new protein complexes. This motivated us to develop the PPISP method for predicting interface residues in protein-protein complexes. In PPISP, sequence profiles and solvent accessibility of spatially neighboring surface residues were used as input to a neural network. The network was trained on native interface residues collected from the Protein Data Bank. The prediction accuracy at the time was 70% with 47% coverage of native interface residues. Now we have extensively improved PPISP. The training set now consisted of 1156 nonhomologous protein chains. Test on a set of 100 nonhomologous protein chains showed that the prediction accuracy is now increased to 80% with 51% coverage. To solve the problem of over-prediction and under-prediction associated with individual neural network models, we developed a consensus method that combines predictions from multiple models with different levels of accuracy and coverage. Applied on a benchmark set of 68 proteins for protein-protein docking, the consensus approach outperformed the best individual models by 3-8 percentage points in accuracy. To demonstrate the predictive power of cons-PPISP, eight complex-forming proteins with interfaces characterized by NMR were tested. These proteins are nonhomologous to the training set and have a total of 144 interface residues identified by chemical shift perturbation. cons-PPISP predicted 174 interface residues with 69% accuracy and 47% coverage and promises to complement experimental techniques in characterizing protein-protein interfaces. .     

Non-covalent interactions across subunit interfaces in Sm proteins

The distinguishing property of Sm protein associations is their high stability. In order to understand this property, we analyzed the interface non-covalent interactions and compared the properties of the Sm protein interfaces with those of a test set, Binding Interface Database (BID). The comparison revealed that the main differences between interfaces of Sm proteins and those of the BID set are the content of charged residues, hydrogen bonds, salt bridges, and conservation scores of interface residues. In Sm proteins, the interfaces have more hydrophobic and fewer charged residues than the surface, which is also the case for the BID test set and other proteins. However, in the interfaces, the content of charged residues in Sm proteins (26%) is substantially larger than that in the BID set (22%). Both interfaces of Sm proteins and of test set have a similar number of hydrophobic interactions per 100 Å². The interfaces of Sm proteins have substantially more hydrogen bonds than the interfaces in test set. The results show clearly that the interfaces of Sm proteins form more salt bridges compared with test set. On average, there are about 16 salt bridges per interface. The high conservation score of amino acids that are involved in non-covalent interactions in protein interfaces is an additional strong argument for their importance. The overriding conclusion from this study is that the non-covalent interactions in Sm protein interfaces considerably contribute to stability of higher order structures.     

Protein subunit interfaces: a statistical analysis of hot spots in Sm proteins

The distinguishing property of Sm protein associations is very high stability. In order to understand this property, we analyzed the interfaces and compared the properties of Sm protein interfaces with those of a test set, the Binding Interface Database (BID). The comparison revealed that the main differences between the interfaces of Sm proteins and those of the BID set are the content of charged residues, the coordination numbers of the residues, knowledge-based pair potentials, and the conservation scores of hot spots. In Sm proteins, the interfaces have more hydrophobic and fewer charged residues than the surfaces, which is also the case for the BID test set and other proteins. However, in the interfaces, the content of charged residues in Sm proteins (26%) is substantially larger than that in the BID set (22%). Hot spots are residues that make up a small fraction of the interfaces, but they contribute most of the binding energy. These residues are critical to protein–protein interactions. Analyses of knowledge-based pair potentials of hot spot and non-hot spot residues in Sm proteins show that they are significantly different; their mean values are 31.5 and 11.3, respectively. In the BID set, this difference is smaller; in this case, the mean values for hot spot and non-hot spot residues are 20.7 and 12.4, respectively. Hence, the pair potentials of hot spots differ significantly for the Sm and BID data sets. In the interfaces of Sm proteins, the amino acids are tightly packed, and the coordination numbers are larger in Sm proteins than in the BID set for both hot spots and non-hot spots. At the same time, the coordination numbers are higher for hot spots; the average coordination number of the hot spot residues in Sm proteins is 7.7, while it is 6.1 for the non-hot spot residues. The difference in the calculated average conservation score for hot spots and non-hot spots in Sm proteins is significantly larger than it is in the BID set. In Sm proteins, the average conservation score for the hot spots is 7.4. Hot spots are surrounded by residues that are moderately conserved (5.9). The average conservation score for the other interface residues is 5.6. The conservation scores in the BID set do not show a significant distinction between hot and non-hot spots: the mean values for hot and non-hot spot residues are 5.5 and 5.2, respectively. These data show that structurally conserved residues and hot spots are significantly correlated in Sm proteins.     

Are protein–protein interfaces more conserved in sequence than the rest of the protein surface?

Protein interfaces are thought to be distinguishable from the rest of the protein surface by their greater degree of residue conservation. We test the validity of this approach on an expanded set of 64 protein-protein interfaces using conservation scores derived from two multiple sequence alignment types, one of close homologs/orthologs and one of diverse homologs/paralogs. Overall, we find that the interface is slightly more conserved than the rest of the protein surface when using either alignment type, with alignments of diverse homologs showing marginally better discrimination. However, using a novel surface-patch definition, we find that the interface is rarely significantly more conserved than other surface patches when using either alignment type. When an interface is among the most conserved surface patches, it tends to be part of an enzyme active site. The most conserved surface patch overlaps with 39% (+/- 28%) and 36% (+/- 28%) of the actual interface for diverse and close homologs, respectively. Contrary to results obtained from smaller data sets, this work indicates that residue conservation is rarely sufficient for complete and accurate prediction of protein interfaces. Finally, we find that obligate interfaces differ from transient interfaces in that the former have significantly fewer alignment gaps at the interface than the rest of the protein surface, as well as having buried interface residues that are more conserved than partially buried interface residues.     

Using protein binding site prediction to improve protein docking

Predicting protein interaction interfaces and protein complexes are two important related problems. For interface prediction, there are a number of tools, such as PPI-Pred, PPISP, PINUP, Promate, and SPPIDER, which predict enzyme-inhibitor interfaces with success rates of 23% to 55% and other interfaces with 10% to 28% on a benchmark dataset of 62 complexes. Here, we develop, metaPPI, a meta server for interface prediction. It significantly improves prediction success rates to 70% for enzyme-inhibitor and 44% for other interfaces. As shown with Promate, predicted interfaces can be used to improve protein docking. Here, we follow this idea using the meta server instead of individual predictions. We confirm that filtering with predicted interfaces significantly improves candidate generation in rigid-body docking based on shape complementarity. Finally, we show that the initial ranking of candidate solutions in rigid-body docking can be further improved for the class of enzyme-inhibitor complexes by a geometrical scoring which rewards deep pockets. A web server of metaPPI is available at scoppi.tu-dresden.de/metappi. The source code of our docking algorithm BDOCK is also available at www.biotec.tu-dresden.de/~bhuang/bdock.     

Development of a protein–ligand-binding site prediction method based on interaction energy and sequence conservation

We present a new method for predicting protein–ligand-binding sites based on protein three-dimensional structure and amino acid conservation. This method involves calculation of the van der Waals interaction energy between a protein and many probes placed on the protein surface and subsequent clustering of the probes with low interaction energies to identify the most energetically favorable locus. In addition, it uses amino acid conservation among homologous proteins. Ligand-binding sites were predicted by combining the interaction energy and the amino acid conservation score. The performance of our prediction method was evaluated using a non-redundant dataset of 348 ligand-bound and ligand-unbound protein structure pairs, constructed by filtering entries in a ligand-binding site structure database, LigASite. Ligand-bound structure prediction (bound prediction) indicated that 74.0 % of predicted ligand-binding sites overlapped with real ligand-binding sites by over 25 % of their volume. Ligand-unbound structure prediction (unbound prediction) indicated that 73.9 % of predicted ligand-binding residues overlapped with real ligand-binding residues. The amino acid conservation score improved the average prediction accuracy by 17.0 and 17.6 points for the bound and unbound predictions, respectively. These results demonstrate the effectiveness of the combined use of the interaction energy and amino acid conservation in the ligand-binding site prediction.     

Residue conservation in viral capsid assembly

To evaluate the evolutionary constraints placed on viral proteins by the structure and assembly of the capsid, we calculate Shannon entropies in the aligned sequences of 45 polypeptide chains in 32 icosahedral viruses, and relate these entropies to the residue location in the three-dimensional structure of the capsids. Three categories of residues have entropies lower than the chain average implying that they are better conserved than average: residues that are buried within a subunit (the protein core), residues that contain atoms buried at an interface between subunits (the interface core), and residues that contribute to several such interfaces. The interface core is also conserved in homomeric proteins and in transient protein-protein complexes, which have only one interface whereas capsids have many. In capsids, the subunit interfaces implicate most of the polypeptide chain: on average, 66% of the capsid residues are at an interface, 34% at more than one, and 47% at the interface core. Nevertheless, we observe that the degree of residue conservation can vary widely between interfaces within a capsid and between regions within an interface. The interfaces and regions of interfaces that show a low sequence variability are likely to play major roles in the self-assembly of the capsid, with implications on its mechanism that we discuss taking adeno-associated virus as an example.     

Prediction of protein interaction sites from sequence profile and residue neighbor list.

Protein-protein interaction sites are predicted from a neural network with sequence profiles of neighboring residues and solvent exposure as input. The network was trained on 615 pairs of nonhomologous complex-forming proteins. Tested on a different set of 129 pairs of nonhomologous complex-forming proteins, 70% of the 11,004 predicted interface residues are actually located in the interfaces. These 7732 correctly predicted residues account for 65% of the 11,805 residues making up the 129 interfaces. The main strength of the network predictor lies in the fact that neighbor lists and solvent exposure are relatively insensitive to structural changes accompanying complex formation. As such, it performs equally well with bound or unbound structures of the proteins. For a set of 35 test proteins, when the input was calculated from the bound and unbound structures, the correct fractions of the predicted interface residues were 69 and 70%, respectively.     

PIER: protein interface recognition for structural proteomics

Recent advances in structural proteomics call for development of fast and reliable automatic methods for prediction of functional surfaces of proteins with known three-dimensional structure, including binding sites for known and unknown protein partners as well as oligomerization interfaces. Despite significant progress the problem is still far from being solved. Most existing methods rely, at least partially, on evolutionary information from multiple sequence alignments projected on protein surface. The common drawback of such methods is their limited applicability to the proteins with a sparse set of sequential homologs, as well as inability to detect interfaces in evolutionary variable regions. In this study, the authors developed an improved method for predicting interfaces from a single protein structure, which is based on local statistical properties of the protein surface derived at the level of atomic groups. The proposed Protein IntErface Recognition (PIER) method achieved the overall precision of 60% at the recall threshold of 50% at the residue level on a diverse benchmark of 490 homodimeric, 62 heterodimeric, and 196 transient interfaces (compared with 25% precision at 50% recall expected from random residue function assignment). For 70% of proteins in the benchmark, the binding patch residues were successfully detected with precision exceeding 50% at 50% recall. The calculation only took seconds for an average 300-residue protein. The authors demonstrated that adding the evolutionary conservation signal only marginally influenced the overall prediction performance on the benchmark; moreover, for certain classes of proteins, using this signal actually resulted in a deteriorated prediction. Thorough benchmarking using other datasets from literature showed that PIER yielded improved performance as compared with several alignment-free or alignment-dependent predictions. The accuracy, efficiency, and dependence on structure alone make PIER a suitable tool for automated high-throughput annotation of protein structures emerging from structural proteomics projects.     

Three-dimensional cluster analysis identifies interfaces and functional residue clusters in proteins

Three-dimensional cluster analysis offers a method for the prediction of functional residue clusters in proteins. This method requires a representative structure and a multiple sequence alignment as input data. Individual residues are represented in terms of regional alignments that reflect both their structural environment and their evolutionary variation, as defined by the alignment of homologous sequences. From the overall (global) and the residue-specific (regional) alignments, we calculate the global and regional similarity matrices, containing scores for all pairwise sequence comparisons in the respective alignments. Comparing the matrices yields two scores for each residue. The regional conservation score (C(R)(x)) defines the conservation of each residue x and its neighbors in 3D space relative to the protein as a whole. The similarity deviation score (S(x)) detects residue clusters with sequence similarities that deviate from the similarities suggested by the full-length sequences. We evaluated 3D cluster analysis on a set of 35 families of proteins with available cocrystal structures, showing small ligand interfaces, nucleic acid interfaces and two types of protein-protein interfaces (transient and stable). We present two examples in detail: fructose-1,6-bisphosphate aldolase and the mitogen-activated protein kinase ERK2. We found that the regional conservation score (C(R)(x)) identifies functional residue clusters better than a scoring scheme that does not take 3D information into account. C(R)(x) is particularly useful for the prediction of poorly conserved, transient protein-protein interfaces. Many of the proteins studied contained residue clusters with elevated similarity deviation scores. These residue clusters correlate with specificity-conferring regions: 3D cluster analysis therefore represents an easily applied method for the prediction of functionally relevant spatial clusters of residues in proteins.     

Statistical analysis and prediction of protein-protein interfaces

Predicting protein-protein interfaces from a three-dimensional structure is a key task of computational structural proteomics. In contrast to geometrically distinct small molecule binding sites, protein-protein interface are notoriously difficult to predict. We generated a large nonredundant data set of 1494 true protein-protein interfaces using biological symmetry annotation where necessary. The data set was carefully analyzed and a Support Vector Machine was trained on a combination of a new robust evolutionary conservation signal with the local surface properties to predict protein-protein interfaces. Fivefold cross validation verifies the high sensitivity and selectivity of the model. As much as 97% of the predicted patches had an overlap with the true interface patch while only 22% of the surface residues were included in an average predicted patch. The model allowed the identification of potential new interfaces and the correction of mislabeled oligomeric states.     

meta-PPISP: a meta web server for protein-protein interaction site prediction

A number of complementary methods have been developed for predicting protein-protein interaction sites. We sought to increase prediction robustness and accuracy by combining results from different predictors, and report here a meta web server, meta-PPISP, that is built on three individual web servers: cons-PPISP (http://pipe.scs.fsu.edu/ppisp.html), Promate (http://bioportal.weizmann.ac.il/promate), and PINUP (http://sparks.informatics.iupui.edu/PINUP/). A linear regression method, using the raw scores of the three servers as input, was trained on a set of 35 nonhomologous proteins. Cross validation showed that meta-PPISP outperforms all the three individual servers. At coverages identical to those of the individual methods, the accuracy of meta-PPISP is higher by 4.8 to 18.2 percentage points. Similar improvements in accuracy are also seen on CAPRI and other targets. AVAILABILITY: meta-PPISP can be accessed at http://pipe.scs.fsu.edu/meta-ppisp.html     

Assessment of protein side‐chain conformation prediction methods in different residue environments

Computational prediction of side‐chain conformation is an important component of protein structure prediction. Accurate side‐chain prediction is crucial for practical applications of protein structure models that need atomic‐detailed resolution such as protein and ligand design. We evaluated the accuracy of eight side‐chain prediction methods in reproducing the side‐chain conformations of experimentally solved structures deposited to the Protein Data Bank. Prediction accuracy was evaluated for a total of four different structural environments (buried, surface, interface, and membrane‐spanning) in three different protein types (monomeric, multimeric, and membrane). Overall, the highest accuracy was observed for buried residues in monomeric and multimeric proteins. Notably, side‐chains at protein interfaces and membrane‐spanning regions were better predicted than surface residues even though the methods did not all use multimeric and membrane proteins for training. Thus, we conclude that the current methods are as practically useful for modeling protein docking interfaces and membrane‐spanning regions as for modeling monomers. Proteins 2014; 82:1971–1984. © 2014 Wiley Periodicals, Inc.     

DISPLAR: an accurate method for predicting DNA-binding sites on protein surfaces

Structural and physical properties of DNA provide important constraints on the binding sites formed on surfaces of DNA-targeting proteins. Characteristics of such binding sites may form the basis for predicting DNA-binding sites from the structures of proteins alone. Such an approach has been successfully developed for predicting protein–protein interface. Here this approach is adapted for predicting DNA-binding sites. We used a representative set of 264 protein–DNA complexes from the Protein Data Bank to analyze characteristics and to train and test a neural network predictor of DNA-binding sites. The input to the predictor consisted of PSI-blast sequence profiles and solvent accessibilities of each surface residue and 14 of its closest neighboring residues. Predicted DNA-contacting residues cover 60% of actual DNA-contacting residues and have an accuracy of 76%. This method significantly outperforms previous attempts of DNA-binding site predictions. Its application to the prion protein yielded a DNA-binding site that is consistent with recent NMR chemical shift perturbation data, suggesting that it can complement experimental techniques in characterizing protein–DNA interfaces.     

Characterization and prediction of the binding site in DNA-binding proteins: improvement of accuracy by combining residue composition, evolutionary conservation and structural parameters

We present a set of four parameters that in combination can predict DNA-binding residues on protein structures to a high degree of accuracy. These are the number of evolutionary conserved residues (N(cons)) and their spatial clustering (ρ(e)), hydrogen bond donor capability (D(p)) and residue propensity (R(p)). We first used these parameters to characterize 130 interfaces in a set of 126 DNA-binding proteins (DBPs). The applicability of these parameters both individually and in combination, to distinguish the true binding region from the rest of the protein surface was then analyzed. R(p) shows the best performance identifying the true interface with the top rank in 83% cases. Importantly, we also used the unbound-bound test cases of the protein-DNA docking benchmark to test the efficacy of our method. When applied to the unbound form of the DBPs, R(p) can distinguish 86% cases. Finally, we have applied the SVM approach for recognizing the interface region using the above parameters along with the individual amino acid composition as attributes. The accuracy of prediction is 90.5% for the bound structures and 93.6% for the unbound form of the proteins.     

Discovery of receptor-ligand interfaces in the immunoglobulin superfamily

Cell-surface-anchored immunoglobulin superfamily (IgSF) proteins are widespread throughout the human proteome, forming crucial components of diverse biological processes including immunity, cell-cell adhesion, and carcinogenesis. IgSF proteins generally function through protein-protein interactions carried out between extracellular, membrane-bound proteins on adjacent cells, known as trans-binding interfaces. These protein-protein interactions constitute a class of pharmaceutical targets important in the treatment of autoimmune diseases, chronic infections, and cancer. A molecular-level understanding of IgSF protein-protein interactions would greatly benefit further drug development. A critical step toward this goal is the reliable identification of IgSF trans-binding interfaces. We propose a novel combination of structure and sequence information to identify trans-binding interfaces in IgSF proteins. We developed a structure-based binding interface prediction approach that can identify broad regions of the protein surface that encompass the binding interfaces and suggests that IgSF proteins possess binding supersites. These interfaces could theoretically be pinpointed using sequence-based conservation analysis, with performance approaching the theoretical upper limit of binding interface prediction accuracy, but achieving this in practice is limited by the current ability to identify an appropriate multiple sequence alignment for conservation analysis. However, an important contribution of combining the two orthogonal methods is that agreement between these approaches can estimate the reliability of the predictions. This approach was benchmarked on the set of 22 IgSF proteins with experimentally solved structures in complex with their ligands. Additionally, we provide structure-based predictions and reliability scores for the 62 IgSF proteins with known structure but yet uncharacterized binding interfaces.     

Residue-level prediction of DNA-binding sites and its application on DNA-binding protein predictions

Protein-DNA interactions are crucial to many cellular activities such as expression-control and DNA-repair. These interactions between amino acids and nucleotides are highly specific and any aberrance at the binding site can render the interaction completely incompetent. In this study, we have three aims focusing on DNA-binding residues on the protein surface: to develop an automated approach for fast and reliable recognition of DNA-binding sites; to improve the prediction by distance-dependent refinement; use these predictions to identify DNA-binding proteins. We use a support vector machines (SVM)-based approach to harness the features of the DNA-binding residues to distinguish them from non-binding residues. Features used for distinction include the residue's identity, charge, solvent accessibility, average potential, the secondary structure it is embedded in, neighboring residues, and location in a cationic patch. These features collected from 50 proteins are used to train SVM. Testing is then performed on another set of 37 proteins, much larger than any testing set used in previous studies. The testing set has no more than 20% sequence identity not only among its pairs, but also with the proteins in the training set, thus removing any undesired redundancy due to homology. This set also has proteins with an unseen DNA-binding structural class not present in the training set. With the above features, an accuracy of 66% with balanced sensitivity and specificity is achieved without relying on homology or evolutionary information. We then develop a post-processing scheme to improve the prediction using the relative location of the predicted residues. Balanced success is then achieved with average sensitivity, specificity and accuracy pegged at 71.3%, 69.3% and 70.5%, respectively. Average net prediction is also around 70%. Finally, we show that the number of predicted DNA-binding residues can be used to differentiate DNA-binding proteins from non-DNA-binding proteins with an accuracy of 78%. Results presented here demonstrate that machine-learning can be applied to automated identification of DNA-binding residues and that the success rate can be ameliorated as more features are added. Such functional site prediction protocols can be useful in guiding consequent works such as site-directed mutagenesis and macromolecular docking.     

Prediction of beta-turns from amino acid sequences using the residue-coupled model.

We evaluated the prediction of beta-turns from amino acid sequences using the residue-coupled model with an enlarged representative protein data set selected from the Protein Data Bank. Our results show that the probability values derived from a data set comprising 425 protein chains yielded an overall beta-turn prediction accuracy 68.74%, compared with 94.7% reported earlier on a data set of 30 proteins using the same method. However, we noted that the overall beta-turn prediction accuracy using probability values derived from the 30-protein data set reduces to 40.74% when tested on the data set comprising 425 protein chains. In contrast, using probability values derived from the 425 data set used in this analysis, the overall beta-turn prediction accuracy yielded consistent results when tested on either the 30-protein data set (64.62%) used earlier or a more recent representative data set comprising 619 protein chains (64.66%) or on a jackknife data set comprising 476 representative protein chains (63.38%). We therefore recommend the use of probability values derived from the 425 representative protein chains data set reported here, which gives more realistic and consistent predictions of beta-turns from amino acid sequences.     

Protein-protein interfaces: analysis of amino acid conservation in homodimers

Evolutionary information derived from the large number of available protein sequences and structures could powerfully guide both analysis and prediction of protein-protein interfaces. To test the relevance of this information, we assess the conservation of residues at protein-protein interfaces compared with other residues on the protein surface. Six homodimer families are analyzed: alkaline phosphatase, enolase, glutathione S-transferase, copper-zinc superoxide dismutase, Streptomyces subtilisin inhibitor, and triose phosphate isomerase. For each family, random simulation is used to calculate the probability (P value) that the level of conservation observed at the interface occurred by chance. The results show that interface conservation is higher than expected by chance and usually statistically significant at the 5% level or better. The effect on the P values of using different definitions of the interface and of excluding active site residues is discussed.     

Predicting DNA-binding amino acid residues from electrostatic stabilization upon mutation to Asp/Glu and evolutionary conservation

Binding of polyanionic DNA depends on the cluster of electropositive atoms in the binding site of a DNA-binding protein. Such a cluster of electropositive protein atoms would be electrostatically unfavorable without stabilizing interactions from the respective electronegative DNA atoms and would likely be evolutionary conserved due to its critical biological role. Consequently, our strategy for predicting DNA-binding residues is based on detecting a cluster of evolutionary conserved surface residues that are electrostatically stabilized upon mutation to negatively charged Asp/Glu residues. The method requires as input the protein structure and sufficient sequence homologs to define each residue's relative conservation, and it yields as output experimentally testable residues that are predicted to bind DNA. By incorporating characteristic DNA-binding site features (i.e., electrostatic strain and amino acid conservation), the new method yields a prediction accuracy of 83%, which is much higher than methods based on only electrostatic strain (57%) or conservation alone (50%). It is also less sensitive to protein conformational changes upon DNA binding than methods that mainly depend on the 3D protein structure.