New insights into the mechanism of protein-protein association

Association of a protein complex follows a two-step mechanism, with the first step being the formation of an encounter complex that evolves into the final complex. Here, we analyze recent experimental data of the association of TEM1-beta-lactamase with BLIP using theoretical calculations and simulation. We show that the calculated Debye-Hückel energy of interaction for a pair of proteins during association resembles an energy funnel, with the final complex at the minima. All attraction is lost at inter-protein distances of 20 A, or rotation angles of >60 degrees from the orientation of the final complex. For faster-associating protein complexes, the energy funnel deepens and its volume increases. Mutations with the largest impact on association (hotspots for association) have the largest effect on the size and depth of the energy funnel. Analyzing existing evidence, we suggest that the transition state along the association pathway is the formation of the final complex from the encounter complex. Consequently, pairs of proteins forming an encounter complex will tend to dissociate more readily than to evolve into the final complex. Increasing directional diffusion by increasing favorable electrostatic attraction results in a faster forming and slower dissociating encounter complex. The possible applicability of electrostatic calculations for protein-protein docking is discussed.     

Electrostatic rate enhancement and transient complex of protein-protein association

The association of two proteins is bounded by the rate at which they, via diffusion, find each other while in appropriate relative orientations. Orientational constraints restrict this rate to approximately 10(5)-10(6) M(-1) s(-1). Proteins with higher association rates generally have complementary electrostatic surfaces; proteins with lower association rates generally are slowed down by conformational changes upon complex formation. Previous studies (Zhou, Biophys J 1997;73:2441-2445) have shown that electrostatic enhancement of the diffusion-limited association rate can be accurately modeled by $k_{\bf D}$ = $k_{D}0\ {exp} ( - \langle U_{el} \rangle;{\star}/k_{B} T),$ where k(D) and k(D0) are the rates in the presence and absence of electrostatic interactions, respectively, U(el) is the average electrostatic interaction energy in a "transient-complex" ensemble, and k(B)T is the thermal energy. The transient-complex ensemble separates the bound state from the unbound state. Predictions of the transient-complex theory on four protein complexes were found to agree well with the experiment when the electrostatic interaction energy was calculated with the linearized Poisson-Boltzmann (PB) equation (Alsallaq and Zhou, Structure 2007;15:215-224). Here we show that the agreement is further improved when the nonlinear PB equation is used. These predictions are obtained with the dielectric boundary defined as the protein van der Waals surface. When the dielectric boundary is instead specified as the molecular surface, electrostatic interactions in the transient complex become repulsive and are thus predicted to retard association. Together these results demonstrate that the transient-complex theory is predictive of electrostatic rate enhancement and can help parameterize PB calculations.     

Prediction-based fingerprints of protein-protein interactions

The recognition of protein interaction sites is an important intermediate step toward identification of functionally relevant residues and understanding protein function, facilitating experimental efforts in that regard. Toward that goal, the authors propose a novel representation for the recognition of protein-protein interaction sites that integrates enhanced relative solvent accessibility (RSA) predictions with high resolution structural data. An observation that RSA predictions are biased toward the level of surface exposure consistent with protein complexes led the authors to investigate the difference between the predicted and actual (i.e., observed in an unbound structure) RSA of an amino acid residue as a fingerprint of interaction sites. The authors demonstrate that RSA prediction-based fingerprints of protein interactions significantly improve the discrimination between interacting and noninteracting sites, compared with evolutionary conservation, physicochemical characteristics, structure-derived and other features considered before. On the basis of these observations, the authors developed a new method for the prediction of protein-protein interaction sites, using machine learning approaches to combine the most informative features into the final predictor. For training and validation, the authors used several large sets of protein complexes and derived from them nonredundant representative chains, with interaction sites mapped from multiple complexes. Alternative machine learning techniques are used, including Support Vector Machines and Neural Networks, so as to evaluate the relative effects of the choice of a representation and a specific learning algorithm. The effects of induced fit and uncertainty of the negative (noninteracting) class assignment are also evaluated. Several representative methods from the literature are reimplemented to enable direct comparison of the results. Using rigorous validation protocols, the authors estimated that the new method yields the overall classification accuracy of about 74% and Matthews correlation coefficients of 0.42, as opposed to up to 70% classification accuracy and up to 0.3 Matthews correlation coefficient for methods that do not utilize RSA prediction-based fingerprints. The new method is available at http://sppider.cchmc.org.     

A novel mammalian display system for the selection of protein-protein interactions by decoy receptor engagement

The emerging field of proteomics has created a need for new high-throughput methodologies for the analysis of gene products. An attractive approach is to develop systems that allow for clonal selection of interacting protein pairs from large molecular libraries. In this study, we have characterized a novel approach for identification and selection of protein-protein interactions, denoted SPIRE (selection of protein interactions by receptor engagement), which is based on a mammalian expression system. We have demonstrated proof of concept by creating a general plasma membrane bound decoy receptor, by displaying a protein or a peptide genetically fused to a trunctated version of the CD40 molecule. When this decoy receptor is engaged by a ligand to the displayed protein/peptide, the receptor expressing cell is rescued from apoptosis. To design a high-throughput system with a highly parallel capacity, we utilized the B cell line WEHI-231, as carrier of the decoy receptor. One specific peptide-displaying cell could be identified and amplified, based on a specific receptor engagement, in a background of 12 500 wild-type cells after four selections. This demonstrates that the approach may serve as a tool in post-genomic research for identifying protein-protein interactions, without prior knowledge of either component.     

Integration of cell-free protein coexpression with an enzyme-linked immunosorbent assay enables rapid analysis of protein-protein interactions directly from DNA

Assays that integrate detection of binding with cell-free protein expression directly from DNA can dramatically increase the pace at which protein-protein interactions (PPIs) can be analyzed by mutagenesis. In this study, we present a method that combines in vitro protein production with an enzyme-linked immunosorbent assay (ELISA) to measure PPIs. This method uses readily available commodity instrumentation and generic antibody-affinity tag interactions. It is straightforward and rapid to execute, enabling many interactions to be assessed in parallel. In traditional ELISAs, reporter complexes are assembled stepwise with one layer at a time. In the method presented here, all the members of the reporter complex are present and assembled together. The signal strength is dependent on all the intercomponent interaction affinities and concentrations. Although this assay is straightforward to execute, establishing proper conditions and analysis of the results require a thorough understanding of the processes that determine the signal strength. The formation of the fully assembled reporter sandwich can be modeled as a competition between Langmuir adsorption isotherms for the immobilized components and binding equilibria of the solution components. We have shown that modeling this process provides semiquantitative understanding of the effects of affinity and concentration and can guide strategies for the development of experimental protocols. We tested the method experimentally using the interaction between a synthetic ankyrin repeat protein (Off7) and maltose-binding protein. Measurements obtained for a collection of alanine mutations in the interface between these two proteins demonstrate that a range of affinities can be analyzed.     

Quantitation of protein-protein interactions by thermal stability shift analysis

Thermal stability shift analysis is a powerful method for examining binding interactions in proteins. We demonstrate that under certain circumstances, protein-protein interactions can be quantitated by monitoring shifts in thermal stability using thermodynamic models and data analysis methods presented in this work. This method relies on the determination of protein stabilities from thermal unfolding experiments using fluorescent dyes such as SYPRO Orange that report on protein denaturation. Data collection is rapid and straightforward using readily available real-time polymerase chain reaction instrumentation. We present an approach for the analysis of the unfolding transitions corresponding to each partner to extract the affinity of the interaction between the proteins. This method does not require the construction of a titration series that brackets the dissociation constant. In thermal shift experiments, protein stability data are obtained at different temperatures according to the affinity- and concentration-dependent shifts in unfolding transition midpoints. Treatment of the temperature dependence of affinity is, therefore, intrinsic to this method and is developed in this study. We used the interaction between maltose-binding protein (MBP) and a thermostable synthetic ankyrin repeat protein (Off7) as an experimental test case because their unfolding transitions overlap minimally. We found that MBP is significantly stabilized by Off7. High experimental throughput is enabled by sample parallelization, and the ability to extract quantitative binding information at a single partner concentration. In a single experiment, we were able to quantify the affinities of a series of alanine mutants, covering a wide range of affinities (～ 100 nM to ～ 100 μM).     

Brownian dynamics of interactions between aldolase mutants and F-actin

Previous Brownian dynamics (BD) simulations (Ouporov IG, Knull HR and Thomasson KA 1999. Biophys. J. 76: 17-27) of complex formation between rabbit aldolase and F-actin have identified three lysine residues (K288, K293 and K341) on aldolase and acidic residues (DEDE) at the N-terminus of actin as important to binding. BD simulations of computer models of aldolase mutants with any of these lysine residues replaced by alanine show reduced binding energy; the greatest effect of a single substitution is for K341A, and replacement of all three lysines greatly reduces binding. BD simulations of wild-type rabbit aldolase vs altered F-actin show that binding is decreased if any one of the four N-terminal acidic residues is replaced by alanine and binding is greatly reduced if three or more of the N-terminal acidic residues are replaced; none of the four actin residues appear more critical for binding than the others.     

Brownian dynamics simulations of glycolytic enzyme subsets with F-actin

Previous Brownian dynamics (BD) simulations identified specific basic residues on fructose-1,6-bisphophate aldolase (aldolase) (I. V. Ouporov et al., Biophysical Journal, 1999, Vol. 76, pp. 17-27) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (I. V. Ouporov et al., Journal of Molecular Recognition, 2001, Vol. 14, pp. 29-41) involved in binding F-actin, and suggested that the quaternary structure of the enzymes may be important. Herein, BD simulations of F-actin binding by enzyme dimers or peptides matching particular sequences of the enzyme and the intact enzyme triose phosphate isomerase (TIM) are compared. BD confirms the experimental observation that TIM has little affinity for F-actin. For aldolase, the critical residues identified by BD are found in surface grooves, formed by subunits A/D and B/C, where they face like residues of the neighboring subunit enhancing their electrostatic potentials. BD simulations between F-actin and aldolase A/D dimers give results similar to the native tetramer. Aldolase A/B dimers form complexes involving residues that are buried in the native structure and are energetically weaker; these results support the importance of quaternary structure for aldolase. GAPDH, however, placed the critical residues on the corners of the tetramer so there is no enhancement of the electrostatic potential between the subunits. Simulations using GAPDH dimers composed of either S/H or G/H subunits show reduced binding energetics compared to the tetramer, but for both dimers, the sets of residues involved in binding are similar to those found for the native tetramer. BD simulations using either aldolase or GAPDH peptides that bind F-actin experimentally show complex formation. The GAPDH peptide bound to the same F-actin domain as did the intact tetramer; however, unlike the tetramer, the aldolase peptide lacked specificity for binding a single F-actin domain.     

Realistic protein-protein association rates from a simple diffusional model neglecting long-range interactions, free energy barriers, and landscape ruggedness

We develop a simple but rigorous model of protein-protein association kinetics based on diffusional association on free energy landscapes obtained by sampling configurations within and surrounding the native complex binding funnels. Guided by results obtained on exactly solvable model problems, we transform the problem of diffusion in a potential into free diffusion in the presence of an absorbing zone spanning the entrance to the binding funnel. The free diffusion problem is solved using a recently derived analytic expression for the rate of association of asymmetrically oriented molecules. Despite the required high steric specificity and the absence of long-range attractive interactions, the computed rates are typically on the order of 10(4)-10(6) M(-1) sec(-1), several orders of magnitude higher than rates obtained using a purely probabilistic model in which the association rate for free diffusion of uniformly reactive molecules is multiplied by the probability of a correct alignment of the two partners in a random collision. As the association rates of many protein-protein complexes are also in the 10(5)-10(6) M(-1) sec(-1) range, our results suggest that free energy barriers arising from desolvation and/or side-chain freezing during complex formation or increased ruggedness within the binding funnel, which are completely neglected in our simple diffusional model, do not contribute significantly to the dynamics of protein-protein association. The transparent physical interpretation of our approach that computes association rates directly from the size and geometry of protein-protein binding funnels makes it a useful complement to Brownian dynamics simulations.     

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. .     

MULTIPROSPECTOR: an algorithm for the prediction of protein-protein interactions by multimeric threading

In this postgenomic era, the ability to identify protein-protein interactions on a genomic scale is very important to assist in the assignment of physiological function. Because of the increasing number of solved structures involving protein complexes, the time is ripe to extend threading to the prediction of quaternary structure. In this spirit, a multimeric threading approach has been developed. The approach is comprised of two phases. In the first phase, traditional threading on a single chain is applied to generate a set of potential structures for the query sequences. In particular, we use our recently developed threading algorithm, PROSPECTOR. Then, for those proteins whose template structures are part of a known complex, we rethread on both partners in the complex and now include a protein-protein interfacial energy. To perform this analysis, a database of multimeric protein structures has been constructed, the necessary interfacial pairwise potentials have been derived, and a set of empirical indicators to identify true multimers based on the threading Z-score and the magnitude of the interfacial energy have been established. The algorithm has been tested on a benchmark set comprised of 40 homodimers, 15 heterodimers, and 69 monomers that were scanned against a protein library of 2478 structures that comprise a representative set of structures in the Protein Data Bank. Of these, the method correctly recognized and assigned 36 homodimers, 15 heterodimers, and 65 monomers. This protocol was applied to identify partners and assign quaternary structures of proteins found in the yeast database of interacting proteins. Our multimeric threading algorithm correctly predicts 144 interacting proteins, compared to the 56 (26) cases assigned by PSI-BLAST using a (less) permissive E-value of 1 (0.01). Next, all possible pairs of yeast proteins have been examined. Predictions (n = 2865) of protein-protein interactions are made; 1138 of these 2865 interactions have counterparts in the Database of Interacting Proteins. In contrast, PSI-BLAST made 1781 predictions, and 1215 have counterparts in DIP. An estimation of the false-negative rate for yeast-predicted interactions has also been provided. Thus, a promising approach to help assist in the assignment of protein-protein interactions on a genomic scale has been developed.     

Cancer-related mutations in BRCA1-BRCT cause long-range structural changes in protein-protein binding sites: a molecular dynamics study

Cancer-associated mutations in the BRCT domain of BRCA1 (BRCA1-BRCT) abolish its tumor suppressor function by disrupting interactions with other proteins such as BACH1. Many cancer-related mutations do not cause sufficient destabilization to lead to global unfolding under physiological conditions, and thus abrogation of function probably is due to localized structural changes. To explore the reasons for mutation-induced loss of function, the authors performed molecular dynamics simulations on three cancer-associated mutants, A1708E, M1775R, and Y1853ter, and on the wild type and benign M1652I mutant, and compared the structures and fluctuations. Only the cancer-associated mutants exhibited significant backbone structure differences from the wild-type crystal structure in BACH1-binding regions, some of which are far from the mutation sites. Backbone differences of the A1708E mutant from the liganded wild type structure in these regions are much larger than those of the unliganded wild type X-ray or molecular dynamics structures. These BACH1-binding regions of the cancer-associated mutants also exhibited increases in their fluctuation magnitudes compared with the same regions in the wild type and M1562I mutant, as quantified by quasiharmonic analysis. Several of the regions of increased fluctuation magnitude correspond to correlated motions of residues in contact that provide a continuous path of fluctuating amino acids in contact from the A1708E and Y1853ter mutation sites to the BACH1-binding sites with altered structure and dynamics. The increased fluctuations in the disease-related mutants suggest an increase in vibrational entropy in the unliganded state that could result in a larger entropy loss in the disease-related mutants upon binding BACH1 than in the wild type. To investigate this possibility, vibrational entropies of the A1708E and wild type in the free state and bound to a BACH1-derived phosphopeptide were calculated using quasiharmonic analysis, to determine the binding entropy difference DeltaDeltaS between the A1708E mutant and the wild type. DeltaDeltaS was determined to be -4.0 cal mol(-1) K(-1), with an uncertainty of 2 cal mol(-1) K(-1); that is, the entropy loss upon binding the peptide is 4.0 cal mol(-1) K(-1) greater for the A1708E mutant, corresponding to an entropic contribution to the DeltaDeltaG of binding (-TDeltaDeltaS) 1.1 kcal mol(-1) more positive for the mutant. The observed differences in structure, flexibility, and entropy of binding likely are responsible for abolition of BACH1 binding, and illustrate that many disease- related mutations could have very long-range effects. The methods described here have potential for identifying correlated motions responsible for other long-range effects of deleterious mutations.     

Cluster conservation as a novel tool for studying protein-protein interactions evolution

Protein-protein interactions networks has come to be a buzzword associated with nets containing edges that represent a pair of interacting proteins (e.g. hormone-receptor, enzyme-inhibitor, antigen-antibody, and a subset of multichain biological machines). Yet, each such interaction composes its own unique network, in which vertices represent amino acid residues, and edges represent atomic contacts. Recent studies have shown that analyses of the data encapsulated in these detailed networks may impact predictions of structure-function correlation. Here, we study homologous families of protein-protein interfaces, which share the same fold but vary in sequence. In this context, we address what properties of the network are shared among relatives with different sequences (and hence different atomic interactions) and which are not. Herein, we develop the general mathematical framework needed to compare the modularity of homologous networks. We then apply this analysis to the structural data of a few interface families, including hemoglobin alpha-beta, growth hormone-receptor, and Serine protease-inhibitor. Our results suggest that interface modularity is an evolutionarily conserved property. Hence, protein-protein interfaces can be clustered down to a few modules, with the boundaries being evolutionarily conserved along homologous complexes. This suggests that protein engineering of protein-protein binding sites may be simplified by varying each module, but retaining the overall modularity of the interface.     

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.     

Temperature differentially affects encounter and docking thermodynamics of antibody--antigen association

Using BIACORE SPR, we have examined the mechanism of temperature effects on the binding kinetics of two closely related antibody Fabs (H10 and H26) which recognize coincident epitopes on hen egg-white lysozyme (HEL), and whose association and dissociation kinetics are best described by the two-step conformational change model which we interpret as molecular encounter and docking. Time-course series data obtained at a series of six temperatures (6, 10, 15, 25, 30 and 37 degrees C) showed that temperature differentially affects the rate constants of the encounter and docking steps. Docking is more temperature-sensitive than the encounter step, and energetically less favorable at higher temperatures. At elevated temperatures, the time required for docking is longer and the apparent increase in off-rate reflects the greater proportion of the molecules failing to dock and remaining in the less stable encounter state. As a consequence, distribution of free energy change between the encounter and docking steps is altered. At physiological temperature (37 degrees C) the docking step of the H26 complex is energetically unfavorable and most complexes essentially do not dock. There is a significant decrease in total free energy change of the H26 complex at higher temperatures. Elevated temperature changes the rate-limiting step of H26--HEL association from the encounter to the docking step, but not that of H10--HEL. Our results indicate that the mechanism by which elevated temperature reduces the affinities of antigen--antibody complexes is to decrease the net docking rate, and/or stability of the docked complex; at higher temperatures, a smaller proportion of the complexes actually anneal to a more stable docked state. This mechanism may have broad applicability to other receptor--ligand complexes.     

Interaction energy decomposition in protein-protein association: a quantum mechanical study of barnase-barstar complex

Protein-protein interactions are very important in the function of a cell. Computational studies of these interactions have been of interest, but often they have utilized classical modelling techniques. In recent years, quantum mechanical (QM) treatment of entire proteins has emerged as a powerful approach to study biomolecular systems. Herein, we apply a semi-empirical divide and conquer (DC) methodology coupled with a dielectric continuum model for the solvent, to explore the contribution of electrostatics, polarization and charge transfer to the interaction energy between barnase and barstar in their complex form. Molecular dynamic (MD) simulation was performed to account for the dynamic behavior of the complex. The results show that electrostatics, charge transfer and polarization favor the formation of the complex. Our study shows that electrostatics dominates the interaction between barnase and barstar ( approximately 73%), while charge transfer and polarization are approximately 21% and approximately 6%, respectively. Close inspection of the polarization and charge-transfer effects on the charge distribution of the complex reveals the existence of two, well localized, regions in barstar. The first region includes the residues between P27 and Y47 and the second region is between N65 and D83. Since no such regions could be detected in barnase clearly suggests that barstar is well optimized for efficiently binding barnase. Furthermore, using our interaction energy decomposition scheme, we were able to identify all residues that have been experimentally determined to be important for the complex formation and to suggest other residues never have been investigated. This suggests that our approach will be useful as an aid in further understanding protein-protein contacts for the ultimate goal to produce successful inhibitors for protein complexes.     

A new, structurally nonredundant, diverse data set of protein-protein interfaces and its implications

Here, we present a diverse, structurally nonredundant data set of two-chain protein-protein interfaces derived from the PDB. Using a sequence order-independent structural comparison algorithm and hierarchical clustering, 3799 interface clusters are obtained. These yield 103 clusters with at least five nonhomologous members. We divide the clusters into three types. In Type I clusters, the global structures of the chains from which the interfaces are derived are also similar. This cluster type is expected because, in general, related proteins associate in similar ways. In Type II, the interfaces are similar; however, remarkably, the overall structures and functions of the chains are different. The functional spectrum is broad, from enzymes/inhibitors to immunoglobulins and toxins. The fact that structurally different monomers associate in similar ways, suggests "good" binding architectures. This observation extends a paradigm in protein science: It has been well known that proteins with similar structures may have different functions. Here, we show that it extends to interfaces. In Type III clusters, only one side of the interface is similar across the cluster. This structurally nonredundant data set provides rich data for studies of protein-protein interactions and recognition, cellular networks and drug design. In particular, it may be useful in addressing the difficult question of what are the favorable ways for proteins to interact. (The data set is available at http://protein3d.ncifcrf.gov/~keskino/ and http://home.ku.edu.tr/~okeskin/INTERFACE/INTERFACES.html.)