Prediction of carbohydrate binding sites on protein surfaces with 3-dimensional probability density distributions of interacting atoms

Non-covalent protein-carbohydrate interactions mediate molecular targeting in many biological processes. Prediction of non-covalent carbohydrate binding sites on protein surfaces not only provides insights into the functions of the query proteins; information on key carbohydrate-binding residues could suggest site-directed mutagenesis experiments, design therapeutics targeting carbohydrate-binding proteins, and provide guidance in engineering protein-carbohydrate interactions. In this work, we show that non-covalent carbohydrate binding sites on protein surfaces can be predicted with relatively high accuracy when the query protein structures are known. The prediction capabilities were based on a novel encoding scheme of the three-dimensional probability density maps describing the distributions of 36 non-covalent interacting atom types around protein surfaces. One machine learning model was trained for each of the 30 protein atom types. The machine learning algorithms predicted tentative carbohydrate binding sites on query proteins by recognizing the characteristic interacting atom distribution patterns specific for carbohydrate binding sites from known protein structures. The prediction results for all protein atom types were integrated into surface patches as tentative carbohydrate binding sites based on normalized prediction confidence level. The prediction capabilities of the predictors were benchmarked by a 10-fold cross validation on 497 non-redundant proteins with known carbohydrate binding sites. The predictors were further tested on an independent test set with 108 proteins. The residue-based Matthews correlation coefficient (MCC) for the independent test was 0.45, with prediction precision and sensitivity (or recall) of 0.45 and 0.49 respectively. In addition, 111 unbound carbohydrate-binding protein structures for which the structures were determined in the absence of the carbohydrate ligands were predicted with the trained predictors. The overall prediction MCC was 0.49. Independent tests on anti-carbohydrate antibodies showed that the carbohydrate antigen binding sites were predicted with comparable accuracy. These results demonstrate that the predictors are among the best in carbohydrate binding site predictions to date.     

A quantitative analysis of interfacial amino acid conservation in protein-protein hetero complexes

A long-standing question in molecular biology is whether interfaces of protein-protein complexes are more conserved than the rest of the protein surfaces. Although it has been reported that conservation can be used as an indicator for predicting interaction sites on proteins, there are recent reports stating that the interface regions are only slightly more conserved than the rest of the protein surfaces, with conservation signals not being statistically significant enough for predicting protein-protein binding sites. In order to properly address these controversial reports we have studied a set of 28 well resolved hetero complex structures of proteins that consists of transient and non-transient complexes. The surface positions were classified into four conservation classes and the conservation index of the surface positions was quantitatively analyzed. The results indicate that the surface density of highly conserved positions is significantly higher in the protein-protein interface regions compared with the other regions of the protein surface. However, the average conservation index of the patches in the interface region is not significantly higher compared with other surface regions of the protein structures. This finding demonstrates that the number of conserved residue positions is a more appropriate indicator for predicting protein-protein binding sites than the average conservation index in the interacting region. We have further validated our findings on a set of 59 benchmark complex structures. Furthermore, an analysis of 19 complexes of antigen-antibody interactions shows that there is no conservation of amino acid positions in the interacting regions of these complexes, as expected, with the variable region of the immunoglobulins interacting mostly with the antigens. Interestingly, antigen interacting regions also have a higher number of non-conserved residue positions in the interacting region than the rest of the protein surface.     

Geometrical and electro-static determinants of protein-protein interactions

Protein-protein interactions (PPI) are pivotal to the numerous processes in the cell. Therefore, it is of interest to document the analysis of these interactions in terms of binding sites, topology of the interacting structures and physiochemical properties of interacting interfaces and the of forces interactions. The interaction interface of obligatory protein-protein complexes differs from that of the transient interactions. We have created a large database of protein-protein interactions containing over100 thousand interfaces. The structural redundancy was eliminated to obtain a non-redundant database of over 2,265 interaction interfaces. Therefore, it is of interest to document the analysis of these interactions in terms of binding sites, topology of the interacting structures and physiochemical properties of interacting interfaces and the offorces interactions. The residue interaction propensity and all of the rest of the parametric scores converged to a statistical indistinguishable common sub-range and followed the similar distribution trends for all three classes of sequence-based classifications PPInS. This indicates that the principles of molecular recognition are dependent on the preciseness of the fit in the interaction interfaces. Thus, it reinforces the importance of geometrical and electrostatic complementarity as the main determinants for PPIs.     

Distribution and complementarity of hydropathy in multisubunit proteins

A survey of 40 multisubunit proteins and 2 protein-protein complexes was performed to assay quantitatively the distribution of hydropathy among the exterior surface, interior, contact surface, and noncontact exterior surface of the isolated subunits. We suggest a useful way to present this distribution by using a "hydropathy level diagram." Additionally, we have devised a function called "hydropathy complementarity" to quantitate the degree to which interacting surfaces have matching hydropathy distributions. Our survey revealed the following patterns: (1) The difference in hydropathy between the interior and exterior of subunits is a fairly invariant quantity. (2) On average, the hydropathy of the contact surface is higher than that of the exterior surface, but is not greater than that of the protein as a whole. There was variation, however, among the proteins. In some instances, the contact surface was more hydrophilic than the noncontact exterior, and in a few cases the contact surface was as hydrophobic as the protein interior. (3) The average interface manifests significant hydropathy complementarity, signifying that proteins interact by placing hydrophobic centers of one surface against hydrophobic centers of the other surface, and by similarly matching hydrophilic centers. As a measure of recognition and specificity, hydropathy complementarity could be a useful tool for predicting correct docking of interacting proteins. We suggest that high hydropathy complementarity is associated with static inflexible interactions. (4) We have found that some subunits that bind predominantly through hydrophilic forces, such as hydrogen bonds, ionic pairs, and water and metal bridges, are involved in dynamic quaternary organization and allostery.     

Can simple codon pair usage predict protein-protein interaction?

Deciphering functional interactions between proteins is one of the great challenges in biology. Sequence-based homology-free encoding schemes have been increasingly applied to develop promising protein-protein interaction (PPI) predictors by means of statistical or machine learning methods. Here we analyze the relationship between codon pair usage and PPIs in yeast. We show that codon pair usage of interacting protein pairs differs significantly from randomly expected. This motivates the development of a novel approach for predicting PPIs, with codon pair frequency difference as input to a Support Vector Machine predictor, termed as CCPPI. 10-fold cross-validation tests based on yeast PPI datasets with balanced positive-to-negative ratios indicate that CCPPI performs better than other sequence-based encoding schemes. Moreover, it ranks the best when tested on an unbalanced large-scale dataset. Although CCPPI is subjected to high false positive rates like many PPI predictors, statistical analyses of the predicted true positives confirm that the success of CCPPI is partly ascribed to its capability to capture proteomic co-expression and functional similarities between interacting protein pairs. Our findings suggest that codon pairs of interacting protein pairs evolve in a coordinated manner and consequently they provide additional information beyond amino acids-based encoding schemes. CCPPI has been made freely available at: http://protein.cau.edu.cn/ccppi.     

Protein complex forming ability is favored over the features of interacting partners in determining the evolutionary rates of proteins in the yeast protein-protein interaction networks

Background  

Evolutionary rates of proteins in a protein-protein interaction network are primarily governed by the protein connectivity and/or expression level. A recent study revealed the importance of the features of the interacting protein partners, viz., the coefficient of functionality and clustering coefficient in controlling the protein evolutionary rates in a protein-protein interaction (PPI) network.     

Fully automated protein complex prediction based on topological similarity and community structure

To understand the function of protein complexes and their association with biological processes, a lot of studies have been done towards analyzing the protein-protein interaction (PPI) networks. However, the advancement in high-throughput technology has resulted in a humongous amount of data for analysis. Moreover, high level of noise, sparseness, and skewness in degree distribution of PPI networks limits the performance of many clustering algorithms and further analysis of their interactions.In addressing and solving these problems we present a novel random walk based algorithm that converts the incomplete and binary PPI network into a protein-protein topological similarity matrix (PP-TS matrix). We believe that if two proteins share some high-order topological similarities they are likely to be interacting with each other. Using the obtained PP-TS matrix, we constructed and used weighted networks to further study and analyze the interaction among proteins. Specifically, we applied a fully automated community structure finding algorithm (Auto-HQcut) on the obtained weighted network to cluster protein complexes. We then analyzed the protein complexes for significance in biological processes. To help visualize and analyze these protein complexes we also developed an interface that displays the resulting complexes as well as the characteristics associated with each complex.Applying our approach to a yeast protein-protein interaction network, we found that the predicted protein-protein interaction pairs with high topological similarities have more significant biological relevance than the original protein-protein interactions pairs. When we compared our PPI network reconstruction algorithm with other existing algorithms using gene ontology and gene co-expression, our algorithm produced the highest similarity scores. Also, our predicted protein complexes showed higher accuracy measure compared to the other protein complex predictions.     

BiGGER: a new (soft) docking algorithm for predicting protein interactions

A new computationally efficient and automated "soft docking" algorithm is described to assist the prediction of the mode of binding between two proteins, using the three-dimensional structures of the unbound molecules. The method is implemented in a software package called BiGGER (Bimolecular Complex Generation with Global Evaluation and Ranking) and works in two sequential steps: first, the complete 6-dimensional binding spaces of both molecules is systematically searched. A population of candidate protein-protein docked geometries is thus generated and selected on the basis of the geometric complementarity and amino acid pairwise affinities between the two molecular surfaces. Most of the conformational changes observed during protein association are treated in an implicit way and test results are equally satisfactory, regardless of starting from the bound or the unbound forms of known structures of the interacting proteins. In contrast to other methods, the entire molecular surfaces are searched during the simulation, using absolutely no additional information regarding the binding sites. In a second step, an interaction scoring function is used to rank the putative docked structures. The function incorporates interaction terms that are thought to be relevant to the stabilization of protein complexes. These include: geometric complementarity of the surfaces, explicit electrostatic interactions, desolvation energy, and pairwise propensities of the amino acid side chains to contact across the molecular interface. The relative functional contribution of each of these interaction terms to the global scoring function has been empirically adjusted through a neural network optimizer using a learning set of 25 protein-protein complexes of known crystallographic structures. In 22 out of 25 protein-protein complexes tested, near-native docked geometries were found with C(alpha) RMS deviations < or =4.0 A from the experimental structures, of which 14 were found within the 20 top ranking solutions. The program works on widely available personal computers and takes 2 to 8 hours of CPU time to run any of the docking tests herein presented. Finally, the value and limitations of the method for the study of macromolecular interactions, not yet revealed by experimental techniques, are discussed.     

Protein-protein interaction as a predictor of subcellular location

Background  

Many biological processes are mediated by dynamic interactions between and among proteins. In order to interact, two proteins must co-occur spatially and temporally. As protein-protein interactions (PPIs) and subcellular location (SCL) are discovered via separate empirical approaches, PPI and SCL annotations are independent and might complement each other in helping us to understand the role of individual proteins in cellular networks. We expect reliable PPI annotations to show that proteins interacting in vivo are co-located in the same cellular compartment. Our goal here is to evaluate the potential of using PPI annotation in determining SCL of proteins in human, mouse, fly and yeast, and to identify and quantify the factors that contribute to this complementarity.     

Self-complementarity within proteins: bridging the gap between binding and folding

Complementarity, in terms of both shape and electrostatic potential, has been quantitatively estimated at protein-protein interfaces and used extensively to predict the specific geometry of association between interacting proteins. In this work, we attempted to place both binding and folding on a common conceptual platform based on complementarity. To that end, we estimated (for the first time to our knowledge) electrostatic complementarity (Em) for residues buried within proteins. Em measures the correlation of surface electrostatic potential at protein interiors. The results show fairly uniform and significant values for all amino acids. Interestingly, hydrophobic side chains also attain appreciable complementarity primarily due to the trajectory of the main chain. Previous work from our laboratory characterized the surface (or shape) complementarity (Sm) of interior residues, and both of these measures have now been combined to derive two scoring functions to identify the native fold amid a set of decoys. These scoring functions are somewhat similar to functions that discriminate among multiple solutions in a protein-protein docking exercise. The performances of both of these functions on state-of-the-art databases were comparable if not better than most currently available scoring functions. Thus, analogously to interfacial residues of protein chains associated (docked) with specific geometry, amino acids found in the native interior have to satisfy fairly stringent constraints in terms of both Sm and Em. The functions were also found to be useful for correctly identifying the same fold for two sequences with low sequence identity. Finally, inspired by the Ramachandran plot, we developed a plot of Sm versus Em (referred to as the complementarity plot) that identifies residues with suboptimal packing and electrostatics which appear to be correlated to coordinate errors.     

Using protein-protein interaction network information to predict the subcellular locations of proteins in budding yeast

The information of protein subcellular localization is vitally important for in-depth understanding the intricate pathways that regulate biological processes at the cellular level. With the rapidly increasing number of newly found protein sequence in the Post-Genomic Age, many automated methods have been developed attempting to help annotate their subcellular locations in a timely manner. However, very few of them were developed using the protein-protein interaction (PPI) network information. In this paper, we have introduced a new concept called "tethering potential" by which the PPI information can be effectively fused into the formulation for protein samples. Based on such a network frame, a new predictor called Yeast-PLoc has been developed for identifying budding yeast proteins among their 19 subcellular location sites. Meanwhile, a purely sequence-based approach, called the "hybrid-property" method, is integrated into Yeast-PLoc as a fall-back to deal with those proteins without sufficient PPI information. The overall success rate by the jackknife test on the 4,683 yeast proteins in the training dataset was 70.25%. Furthermore, it was shown that the success rate by Yeast- PLoc on an independent dataset was remarkably higher than those by some other existing predictors, indicating that the current approach by incorporating the PPI information is quite promising. As a user-friendly web-server, Yeast-PLoc is freely accessible at http://yeastloc.biosino.org/.     

Prediction of protein-protein interactions using protein signature profiling

Protein domains are conserved and functionally independent structures that play an important role in interactions among related proteins. Domain-domain interactions have been recently used to predict protein-protein interactions (PPI). In general, the interaction probability of a pair of domains is scored using a trained scoring function. Satisfying a threshold, the protein pairs carrying those domains are regarded as "interacting". In this study, the signature contents of proteins were utilized to predict PPI pairs in Saccharomyces cerevisiae, Caenorhabditis elegans, and Homo sapiens. Similarity between protein signature patterns was scored and PPI predictions were drawn based on the binary similarity scoring function. Results show that the true positive rate of prediction by the proposed approach is approximately 32% higher than that using the maximum likelihood estimation method when compared with a test set, resulting in 22% increase in the area under the receiver operating characteristic (ROC) curve. When proteins containing one or two signatures were removed, the sensitivity of the predicted PPI pairs increased significantly. The predicted PPI pairs are on average 11 times more likely to interact than the random selection at a confidence level of 0.95, and on average 4 times better than those predicted by either phylogenetic profiling or gene expression profiling.     

Comparative analysis of differential proteome-wide protein-protein interaction network of Methanobrevibacter ruminantium M1

A proteome-wide protein-protein interaction (PPI) network of Methanobrevibacter ruminantium M1 (MRU), a predominant rumen methanogen, was constructed from its metabolic genes using a gene neighborhood algorithm and then compared with closely related rumen methanogens Using proteome-wide PPI approach, we constructed network encompassed 2194 edges and 637 nodes interacting with 634 genes. Network quality and robustness of functional modules were assessed with gene ontology terms. A structure-function-metabolism mapping for each protein has been carried out with efforts to extract experimental PPI concomitant information from the literature. The results of our study revealed that some topological properties of its network were robust for sharing homologous protein interactions across heterotrophic and hydrogenotrophic methanogens. MRU proteome has shown to establish many PPI sub-networks for associated metabolic subsystems required to survive in the rumen environment. MRU genome found to share interacting proteins from its PPI network involved in specific metabolic subsystems distinct to heterotrophic and hydrogenotrophic methanogens. Across these proteomes, the interacting proteins from differential PPI networks were shared in common for the biosynthesis of amino acids, nucleosides, and nucleotides and energy metabolism in which more fractions of protein pairs shared with Methanosarcina acetivorans. Our comparative study expedites our knowledge to understand a complex proteome network associated with typical metabolic subsystems of MRU and to improve its genome-scale reconstruction in the future.     

Predicting Human Protein Subcellular Locations by the Ensemble of Multiple Predictors via Protein-Protein Interaction Network with Edge Clustering Coefficients

One of the fundamental tasks in biology is to identify the functions of all proteins to reveal the primary machinery of a cell. Knowledge of the subcellular locations of proteins will provide key hints to reveal their functions and to understand the intricate pathways that regulate biological processes at the cellular level. Protein subcellular location prediction has been extensively studied in the past two decades. A lot of methods have been developed based on protein primary sequences as well as protein-protein interaction network. In this paper, we propose to use the protein-protein interaction network as an infrastructure to integrate existing sequence based predictors. When predicting the subcellular locations of a given protein, not only the protein itself, but also all its interacting partners were considered. Unlike existing methods, our method requires neither the comprehensive knowledge of the protein-protein interaction network nor the experimentally annotated subcellular locations of most proteins in the protein-protein interaction network. Besides, our method can be used as a framework to integrate multiple predictors. Our method achieved 56% on human proteome in absolute-true rate, which is higher than the state-of-the-art methods.     

结构域相互作用数据库的产生、发展与应用

结构域是进化上的保守序列单元,是蛋白质的结构和功能的标准组件．典型的两个蛋白质间的相互作用涉及特殊结构域间的结合,而且识别相互作用结构域对于在结构域水平上彻底理解蛋白质的功能与进化、构建蛋白质相互作用网络、分析生物学通路等十分重要．目前,依赖于对实验数据的进一步挖掘和对各种不同输入数据的计算预测,已识别出了一些相互作用/功能连锁结构域对,并由此构建了内容丰富、日益更新的结构域相互作用数据库．综述了产生结构域相互作用的8种计算预测方法．介绍了5个结构域相互作用公共数据库3DID、iPfam、InterDom、DIMA和DOMINE的有关信息和最新动态．实例概述了结构域相互作用在蛋白质相互作用计算预测、可信度评估,蛋白质结构域注释,以及在生物学通路分析中的应用．     

Prediction of Protein-Protein Interactions Using Protein Signature Profiling

Protein domains are conserved and functionally independent structures that play an important role in interactions among related proteins. Domain-domain inter- actions have been recently used to predict protein-protein interactions (PPI). In general, the interaction probability of a pair of domains is scored using a trained scoring function. Satisfying a threshold, the protein pairs carrying those domains are regarded as "interacting". In this study, the signature contents of proteins were utilized to predict PPI pairs in Saccharomyces cerevisiae, Caenorhabditis ele- gans, and Homo sapiens. Similarity between protein signature patterns was scored and PPI predictions were drawn based on the binary similarity scoring function. Results show that the true positive rate of prediction by the proposed approach is approximately 32% higher than that using the maximum likelihood estimation method when compared with a test set, resulting in 22% increase in the area un- der the receiver operating characteristic (ROC) curve. When proteins containing one or two signatures were removed, the sensitivity of the predicted PPI pairs in- creased significantly. The predicted PPI pairs are on average 11 times more likely to interact than the random selection at a confidence level of 0.95, and on aver- age 4 times better than those predicted by either phylogenetic profiling or gene expression profiling.     

Identification of Essential Proteins Based on Ranking Edge-Weights in Protein-Protein Interaction Networks

Essential proteins are those that are indispensable to cellular survival and development. Existing methods for essential protein identification generally rely on knock-out experiments and/or the relative density of their interactions (edges) with other proteins in a Protein-Protein Interaction (PPI) network. Here, we present a computational method, called EW, to first rank protein-protein interactions in terms of their Edge Weights, and then identify sub-PPI-networks consisting of only the highly-ranked edges and predict their proteins as essential proteins. We have applied this method to publicly-available PPI data on Saccharomyces cerevisiae (Yeast) and Escherichia coli (E. coli) for essential protein identification, and demonstrated that EW achieves better performance than the state-of-the-art methods in terms of the precision-recall and Jackknife measures. The highly-ranked protein-protein interactions by our prediction tend to be biologically significant in both the Yeast and E. coli PPI networks. Further analyses on systematically perturbed Yeast and E. coli PPI networks through randomly deleting edges demonstrate that the proposed method is robust and the top-ranked edges tend to be more associated with known essential proteins than the lowly-ranked edges.     

Shape complementarity of protein-protein complexes at multiple resolutions

Biological complexes typically exhibit intermolecular interfaces of high shape complementarity. Many computational docking approaches use this surface complementarity as a guide in the search for predicting the structures of protein-protein complexes. Proteins often undergo conformational changes to create a highly complementary interface when associating. These conformational changes are a major cause of failure for automated docking procedures when predicting binding modes between proteins using their unbound conformations. Low resolution surfaces in which high frequency geometric details are omitted have been used to address this problem. These smoothed, or blurred, surfaces are expected to minimize the differences between free and bound structures, especially those that are due to side chain conformations or small backbone deviations. Despite the fact that this approach has been used in many docking protocols, there has yet to be a systematic study of the effects of such surface smoothing on the shape complementarity of the resulting interfaces. Here we investigate this question by computing shape complementarity of a set of 66 protein-protein complexes represented by multiresolution blurred surfaces. Complexed and unbound structures are available for these protein-protein complexes. They are a subset of complexes from a nonredundant docking benchmark selected for rigidity (i.e. the proteins undergo limited conformational changes between their bound and unbound states). In this work, we construct the surfaces by isocontouring a density map obtained by accumulating the densities of Gaussian functions placed at all atom centers of the molecule. The smoothness or resolution is specified by a Gaussian fall-off coefficient, termed "blobbyness." Shape complementarity is quantified using a histogram of the shortest distances between two proteins' surface mesh vertices for both the crystallographic complexes and the complexes built using the protein structures in their unbound conformation. The histograms calculated for the bound complex structures demonstrate that medium resolution smoothing (blobbyness = -0.9) can reproduce about 88% of the shape complementarity of atomic resolution surfaces. Complexes formed from the free component structures show a partial loss of shape complementarity (more overlaps and gaps) with the atomic resolution surfaces. For surfaces smoothed to low resolution (blobbyness = -0.3), we find more consistency of shape complementarity between the complexed and free cases. To further reduce bad contacts without significantly impacting the good contacts we introduce another blurred surface, in which the Gaussian densities of flexible atoms are reduced. From these results we discuss the use of shape complementarity in protein-protein docking.     

Alignment of non-covalent interactions at protein-protein interfaces

Background

The study and comparison of protein-protein interfaces is essential for the understanding of the mechanisms of interaction between proteins. While there are many methods for comparing protein structures and protein binding sites, so far no methods have been reported for comparing the geometry of non-covalent interactions occurring at protein-protein interfaces.

Methodology/Principal Findings

Here we present a method for aligning non-covalent interactions between different protein-protein interfaces. The method aligns the vector representations of van der Waals interactions and hydrogen bonds based on their geometry. The method has been applied to a dataset which comprises a variety of protein-protein interfaces. The alignments are consistent to a large extent with the results obtained using two other complementary approaches. In addition, we apply the method to three examples of protein mimicry. The method successfully aligns respective interfaces and allows for recognizing conserved interface regions.

Conclusions/Significance

The Galinter method has been validated in the comparison of interfaces in which homologous subunits are involved, including cases of mimicry. The method is also applicable to comparing interfaces involving non-peptidic compounds. Galinter assists users in identifying local interface regions with similar patterns of non-covalent interactions. This is particularly relevant to the investigation of the molecular basis of interaction mimicry.