SH3-SPOT: an algorithm to predict preferred ligands to different members of the SH3 gene family

Incorporating physics to overcome data scarcity in predictive modeling of protein function: A case study of BK channels

Machine learning has played transformative roles in numerous chemical and biophysical problems such as protein folding where large amount of data exists. Nonetheless, many important problems remain challenging for data-driven machine learning approaches due to the limitation of data scarcity. One approach to overcome data scarcity is to incorporate physical principles such as through molecular modeling and simulation. Here, we focus on the big potassium (BK) channels that play important roles in cardiovascular and neural systems. Many mutants of BK channel are associated with various neurological and cardiovascular diseases, but the molecular effects are unknown. The voltage gating properties of BK channels have been characterized for 473 site-specific mutations experimentally over the last three decades; yet, these functional data by themselves remain far too sparse to derive a predictive model of BK channel voltage gating. Using physics-based modeling, we quantify the energetic effects of all single mutations on both open and closed states of the channel. Together with dynamic properties derived from atomistic simulations, these physical descriptors allow the training of random forest models that could reproduce unseen experimentally measured shifts in gating voltage, ∆V1/2, with a RMSE ~ 32 mV and correlation coefficient of R ~ 0.7. Importantly, the model appears capable of uncovering nontrivial physical principles underlying the gating of the channel, including a central role of hydrophobic gating. The model was further evaluated using four novel mutations of L235 and V236 on the S5 helix, mutations of which are predicted to have opposing effects on V1/2 and suggest a key role of S5 in mediating voltage sensor-pore coupling. The measured ∆V1/2 agree quantitatively with prediction for all four mutations, with a high correlation of R = 0.92 and RMSE = 18 mV. Therefore, the model can capture nontrivial voltage gating properties in regions where few mutations are known. The success of predictive modeling of BK voltage gating demonstrates the potential of combining physics and statistical learning for overcoming data scarcity in nontrivial protein function prediction.

Incorporating physics to overcome data scarcity in predictive modeling of protein function: a case study of BK channels

Machine learning has played transformative roles in numerous chemical and biophysical problems such as protein folding where large amount of data exists. Nonetheless, many important problems remain challenging for data-driven machine learning approaches due to the limitation of data scarcity. One approach to overcome data scarcity is to incorporate physical principles such as through molecular modeling and simulation. Here, we focus on the big potassium (BK) channels that play important roles in cardiovascular and neural systems. Many mutants of BK channel are associated with various neurological and cardiovascular diseases, but the molecular effects are unknown. The voltage gating properties of BK channels have been characterized for 473 site-specific mutations experimentally over the last three decades; yet, these functional data by themselves remain far too sparse to derive a predictive model of BK channel voltage gating. Using physics-based modeling, we quantify the energetic effects of all single mutations on both open and closed states of the channel. Together with dynamic properties derived from atomistic simulations, these physical descriptors allow the training of random forest models that could reproduce unseen experimentally measured shifts in gating voltage, ΔV 1/2 , with a RMSE ∼ 32 mV and correlation coefficient of R ∼ 0.7. Importantly, the model appears capable of uncovering nontrivial physical principles underlying the gating of the channel, including a central role of hydrophobic gating. The model was further evaluated using four novel mutations of L235 and V236 on the S5 helix, mutations of which are predicted to have opposing effects on V 1/2 and suggest a key role of S5 in mediating voltage sensor-pore coupling. The measured ΔV 1/2 agree quantitatively with prediction for all four mutations, with a high correlation of R = 0.92 and RMSE = 18 mV. Therefore, the model can capture nontrivial voltage gating properties in regions where few mutations are known. The success of predictive modeling of BK voltage gating demonstrates the potential of combining physics and statistical learning for overcoming data scarcity in nontrivial protein function prediction.

Author Summary

Deep machine learning has brought many exciting breakthroughs in chemistry, physics and biology. These models require large amount of training data and struggle when the data is scarce. The latter is true for predictive modeling of the function of complex proteins such as ion channels, where only hundreds of mutational data may be available. Using the big potassium (BK) channel as a biologically important model system, we demonstrate that a reliable predictive model of its voltage gating property could be derived from only 473 mutational data by incorporating physics-derived features, which include dynamic properties from molecular dynamics simulations and energetic quantities from Rosetta mutation calculations. We show that the final random forest model captures key trends and hotspots in mutational effects of BK voltage gating, such as the important role of pore hydrophobicity. A particularly curious prediction is that mutations of two adjacent residues on the S5 helix would always have opposite effects on the gating voltage, which was confirmed by experimental characterization of four novel mutations. The current work demonstrates the importance and effectiveness of incorporating physics in predictive modeling of protein function with scarce data.

Computationally-Aided Modeling of Hsp70-Client Interactions: Past, Present, and Future

Hsp70 molecular chaperones play central roles in maintaining a healthy cellular proteome. Hsp70s function by binding to short peptide sequences in incompletely folded client proteins, thus preventing them from misfolding and/or aggregating, and in many cases holding them in a state that is competent for subsequent processes like translocation across membranes. There is considerable interest in predicting the sites where Hsp70s may bind their clients, as the ability to do so sheds light on the cellular functions of the chaperone. In addition, the capacity of the Hsp70 chaperone family to bind to a broad array of clients and to identify accessible sequences that enable discrimination of those that are folded from those that are not fully folded, which is essential to their cellular roles, is a fascinating puzzle in molecular recognition. In this article we discuss efforts to harness computational modeling with input from experimental data to develop a predictive understanding of the promiscuous yet selective binding of Hsp70 molecular chaperones to accessible sequences within their client proteins. We trace how an increasing understanding of the complexities of Hsp70-client interactions has led computational modeling to new underlying assumptions and design features. We describe the trend from purely data-driven analysis toward increased reliance on physics-based modeling that deeply integrates structural information and sequence-based functional data with physics-based binding energies. Notably, new experimental insights are adding to our understanding of the molecular origins of "selective promiscuity" in substrate binding by Hsp70 chaperones and challenging the underlying assumptions and design used in earlier predictive models. Taking the new experimental findings together with exciting progress in computational modeling of protein structures leads us to foresee a bright future for a predictive understanding of selective-yet-promiscuous binding exploited by Hsp70 molecular chaperones; the resulting new insights will also apply to substrate binding by other chaperones and by signaling proteins.

Understanding binding affinity and specificity of modular protein domains: A focus in ligand design for the polyproline-binding families

Within the modular protein domains there are five families that recognize proline-rich sequences: SH3, WW, EVH1, GYF and UEV domains. This chapter reviews the main strategies developed for the design of ligands for these families, including peptides, peptidomimetics and drugs. We also describe some studies aimed to understand the molecular reasons responsible for the intrinsic affinity and specificity of these domains.

Large-Scale Screening of Preferred Interactions of Human Src Homology-3 (SH3) Domains Using Native Target Proteins as Affinity Ligands

The Src Homology-3 (SH3) domains are ubiquitous protein modules that mediate important intracellular protein interactions via binding to short proline-rich consensus motifs in their target proteins. The affinity and specificity of such core SH3 - ligand contacts are typically modest, but additional binding interfaces can give rise to stronger and more specific SH3-mediated interactions. To understand how commonly such robust SH3 interactions occur in the human protein interactome, and to identify these in an unbiased manner we have expressed 324 predicted human SH3 ligands as full-length proteins in mammalian cells, and screened for their preferred SH3 partners using a phage display-based approach. This discovery platform contains an essentially complete repertoire of the ∼300 human SH3 domains, and involves an inherent binding threshold that ensures selective identification of only SH3 interactions with relatively high affinity. Such strong and selective SH3 partners could be identified for only 19 of these 324 predicted ligand proteins, suggesting that the majority of human SH3 interactions are relatively weak, and thereby have capacity for only modest inherent selectivity. The panel of exceptionally robust SH3 interactions identified here provides a rich source of leads and hypotheses for further studies. However, a truly comprehensive characterization of the human SH3 interactome will require novel high-throughput methods based on function instead of absolute binding affinity.

Learning sequence determinants of protein:protein interaction specificity with sparse graphical models

In studying the strength and specificity of interaction between members of two protein families, key questions center on which pairs of possible partners actually interact, how well they interact, and why they interact while others do not. The advent of large-scale experimental studies of interactions between members of a target family and a diverse set of possible interaction partners offers the opportunity to address these questions. We develop here a method, DgSpi (data-driven graphical models of specificity in protein:protein interactions), for learning and using graphical models that explicitly represent the amino acid basis for interaction specificity (why) and extend earlier classification-oriented approaches (which) to predict the ΔG of binding (how well). We demonstrate the effectiveness of our approach in analyzing and predicting interactions between a set of 82 PDZ recognition modules against a panel of 217 possible peptide partners, based on data from MacBeath and colleagues. Our predicted ΔG values are highly predictive of the experimentally measured ones, reaching correlation coefficients of 0.69 in 10-fold cross-validation and 0.63 in leave-one-PDZ-out cross-validation. Furthermore, the model serves as a compact representation of amino acid constraints underlying the interactions, enabling protein-level ΔG predictions to be naturally understood in terms of residue-level constraints. Finally, the model DgSpi readily enables the design of new interacting partners, and we demonstrate that designed ligands are novel and diverse.

MoDPepInt: an interactive web server for prediction of modular domain-peptide interactions

Summary: MoDPepInt (Modular Domain Peptide Interaction) is a new easy-to-use web server for the prediction of binding partners for modular protein domains. Currently, we offer models for SH2, SH3 and PDZ domains via the tools SH2PepInt, SH3PepInt and PDZPepInt, respectively. More specifically, our server offers predictions for 51 SH2 human domains and 69 SH3 human domains via single domain models, and predictions for 226 PDZ domains across several species, via 43 multidomain models. All models are based on support vector machines with different kernel functions ranging from polynomial, to Gaussian, to advanced graph kernels. In this way, we model non-linear interactions between amino acid residues. Results were validated on manually curated datasets achieving competitive performance against various state-of-the-art approaches.

Availability and implementation: The MoDPepInt server is available under the URL http://modpepint.informatik.uni-freiburg.de/

Contact: backofen@informatik.uni-freiburg.de

Supplementary information: Supplementary data are available at Bioinformatics online.

Learning Sequence Determinants of Protein:protein Interaction Specificity with Sparse Graphical Models

In studying the strength and specificity of interaction between members of two protein families, key questions center on which pairs of possible partners actually interact, how well they interact, and why they interact while others do not. The advent of large-scale experimental studies of interactions between members of a target family and a diverse set of possible interaction partners offers the opportunity to address these questions. We develop here a method, DgSpi (Data-driven Graphical models of Specificity in Protein:protein Interactions), for learning and using graphical models that explicitly represent the amino acid basis for interaction specificity (why) and extend earlier classification-oriented approaches (which) to predict the ΔG of binding (how well). We demonstrate the effectiveness of our approach in analyzing and predicting interactions between a set of 82 PDZ recognition modules, against a panel of 217 possible peptide partners, based on data from MacBeath and colleagues. Our predicted ΔG values are highly predictive of the experimentally measured ones, reaching correlation coefficients of 0.69 in 10-fold cross-validation and 0.63 in leave-one-PDZ-out cross-validation. Furthermore, the model serves as a compact representation of amino acid constraints underlying the interactions, enabling protein-level ΔG predictions to be naturally understood in terms of residue-level constraints. Finally, as a generative model, DgSpi readily enables the design of new interacting partners, and we demonstrate that designed ligands are novel and diverse.

A graph kernel approach for alignment-free domain-peptide interaction prediction with an application to human SH3 domains

MOTIVATION

State-of-the-art experimental data for determining binding specificities of peptide recognition modules (PRMs) is obtained by high-throughput approaches like peptide arrays. Most prediction tools applicable to this kind of data are based on an initial multiple alignment of the peptide ligands. Building an initial alignment can be error-prone, especially in the case of the proline-rich peptides bound by the SH3 domains.

RESULTS

Here, we present a machine-learning approach based on an efficient graph-kernel technique to predict the specificity of a large set of 70 human SH3 domains, which are an important class of PRMs. The graph-kernel strategy allows us to (i) integrate several types of physico-chemical information for each amino acid, (ii) consider high-order correlations between these features and (iii) eliminate the need for an initial peptide alignment. We build specialized models for each human SH3 domain and achieve competitive predictive performance of 0.73 area under precision-recall curve, compared with 0.27 area under precision-recall curve for state-of-the-art methods based on position weight matrices. We show that better models can be obtained when we use information on the noninteracting peptides (negative examples), which is currently not used by the state-of-the art approaches based on position weight matrices. To this end, we analyze two strategies to identify subsets of high confidence negative data. The techniques introduced here are more general and hence can also be used for any other protein domains, which interact with short peptides (i.e. other PRMs).

AVAILABILITY

The program with the predictive models can be found at http://www.bioinf.uni-freiburg.de/Software/SH3PepInt/SH3PepInt.tar.gz. We also provide a genome-wide prediction for all 70 human SH3 domains, which can be found under http://www.bioinf.uni-freiburg.de/Software/SH3PepInt/Genome-Wide-Predictions.tar.gz.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Identification of methyllysine peptides binding to chromobox protein homolog 6 chromodomain in the human proteome

Methylation is one of the important post-translational modifications that play critical roles in regulating protein functions. Proteomic identification of this post-translational modification and understanding how it affects protein activity remain great challenges. We tackled this problem from the aspect of methylation mediating protein-protein interaction. Using the chromodomain of human chromobox protein homolog 6 as a model system, we developed a systematic approach that integrates structure modeling, bioinformatics analysis, and peptide microarray experiments to identify lysine residues that are methylated and recognized by the chromodomain in the human proteome. Given the important role of chromobox protein homolog 6 as a reader of histone modifications, it was interesting to find that the majority of its interacting partners identified via this approach function in chromatin remodeling and transcriptional regulation. Our study not only illustrates a novel angle for identifying methyllysines on a proteome-wide scale and elucidating their potential roles in regulating protein function, but also suggests possible strategies for engineering the chromodomain-peptide interface to enhance the recognition of and manipulate the signal transduction mediated by such interactions.

Prediction of contact matrix for protein-protein interaction

MOTIVATION

Prediction of protein-protein interaction has become an important part of systems biology in reverse engineering the biological networks for better understanding the molecular biology of the cell. Although significant progress has been made in terms of prediction accuracy, most computational methods only predict whether two proteins interact but not their interacting residues-the information that can be very valuable for understanding the interaction mechanisms and designing modulation of the interaction. In this work, we developed a computational method to predict the interacting residue pairs-contact matrix for interacting protein domains, whose rows and columns correspond to the residues in the two interacting domains respectively and whose values (1 or 0) indicate whether the corresponding residues (do or do not) interact.

RESULTS

Our method is based on supervised learning using support vector machines. For each domain involved in a given domain-domain interaction (DDI), an interaction profile hidden Markov model (ipHMM) is first built for the domain family, and then each residue position for a member domain sequence is represented as a 20-dimension vector of Fisher scores, characterizing how similar it is as compared with the family profile at that position. Each element of the contact matrix for a sequence pair is now represented by a feature vector from concatenating the vectors of the two corresponding residues, and the task is to predict the element value (1 or 0) from the feature vector. A support vector machine is trained for a given DDI, using either a consensus contact matrix or contact matrices for individual sequence pairs, and is tested by leave-one-out cross validation. The performance averaged over a set of 115 DDIs collected from the 3 DID database shows significant improvement (sensitivity up to 85%, and specificity up to 85%), as compared with a multiple sequence alignment-based method (sensitivity 57%, and specificity 78%) previously reported in the literature.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Characterization of domain-peptide interaction interface: prediction of SH3 domain-mediated protein-protein interaction network in yeast by generic structure-based models

Determination of the binding specificity of SH3 domain, a peptide recognition module (PRM), is important to understand their biological functions and reconstruct the SH3-mediated protein-protein interaction network. In the present study, the SH3-peptide interactions for both class I and II SH3 domains were characterized by the intermolecular residue-residue interaction network. We developed generic MIEC-SVM models to infer SH3 domain-peptide recognition specificity that achieved satisfactory prediction accuracy. By investigating the domain-peptide recognition mechanisms at the residue level, we found that the class-I and class-II binding peptides have different binding modes even though they occupy the same binding site of SH3. Furthermore, we predicted the potential binding partners of SH3 domains in the yeast proteome and constructed the SH3-mediated protein-protein interaction network. Comparison with the experimentally determined interactions confirmed the effectiveness of our approach. This study showed that our sophisticated computational approach not only provides a powerful platform to decipher protein recognition code at the molecular level but also allows identification of peptide-mediated protein interactions at a proteomic scale. We believe that such an approach is general to be applicable to other domain-peptide interactions.

DomPep--a general method for predicting modular domain-mediated protein-protein interactions

Protein-protein interactions (PPIs) are frequently mediated by the binding of a modular domain in one protein to a short, linear peptide motif in its partner. The advent of proteomic methods such as peptide and protein arrays has led to the accumulation of a wealth of interaction data for modular interaction domains. Although several computational programs have been developed to predict modular domain-mediated PPI events, they are often restricted to a given domain type. We describe DomPep, a method that can potentially be used to predict PPIs mediated by any modular domains. DomPep combines proteomic data with sequence information to achieve high accuracy and high coverage in PPI prediction. Proteomic binding data were employed to determine a simple yet novel parameter Ligand-Binding Similarity which, in turn, is used to calibrate Domain Sequence Identity and Position-Weighted-Matrix distance, two parameters that are used in constructing prediction models. Moreover, DomPep can be used to predict PPIs for both domains with experimental binding data and those without. Using the PDZ and SH2 domain families as test cases, we show that DomPep can predict PPIs with accuracies superior to existing methods. To evaluate DomPep as a discovery tool, we deployed DomPep to identify interactions mediated by three human PDZ domains. Subsequent in-solution binding assays validated the high accuracy of DomPep in predicting authentic PPIs at the proteome scale. Because DomPep makes use of only interaction data and the primary sequence of a domain, it can be readily expanded to include other types of modular domains.

Prediction of peptides binding to the PKA RIIalpha subunit using a hierarchical strategy

MOTIVATION

Favorable interaction between the regulatory subunit of the cAMP-dependent protein kinase (PKA) and a peptide in A-kinase anchoring proteins (AKAPs) is critical for translocating PKA to the subcellular sites where the enzyme phosphorylates its substrates. It is very hard to identify AKAPs peptides binding to PKA due to the high sequence diversity of AKAPs.

RESULTS

We propose a hierarchical and efficient approach, which combines molecular dynamics (MD) simulations, free energy calculations, virtual mutagenesis (VM) and bioinformatics analyses, to predict peptides binding to the PKA RIIα regulatory subunit in the human proteome systematically. Our approach successfully retrieved 15 out of 18 documented RIIα-binding peptides. Literature curation supported that many newly predicted peptides might be true AKAPs. Here, we present the first systematic search for AKAP peptides in the human proteome, which is useful to further experimental identification of AKAPs and functional analysis of their biological roles.

Binding of the cSH3 domain of Grb2 adaptor to two distinct RXXK motifs within Gab1 docker employs differential mechanisms

A ubiquitous component of cellular signaling machinery, Gab1 docker plays a pivotal role in routing extracellular information in the form of growth factors and cytokines to downstream targets such as transcription factors within the nucleus. Here, using isothermal titration calorimetry (ITC) in combination with macromolecular modeling (MM), we show that although Gab1 contains four distinct RXXK motifs, designated G1, G2, G3, and G4, only G1 and G2 motifs bind to the cSH3 domain of Grb2 adaptor and do so with distinct mechanisms. Thus, while the G1 motif strictly requires the PPRPPKP consensus sequence for high-affinity binding to the cSH3 domain, the G2 motif displays preference for the PXVXRXLKPXR consensus. Such sequential differences in the binding of G1 and G2 motifs arise from their ability to adopt distinct polyproline type II (PPII)- and 3(10) -helical conformations upon binding to the cSH3 domain, respectively. Collectively, our study provides detailed biophysical insights into a key protein-protein interaction involved in a diverse array of signaling cascades central to health and disease.

SH3 domain-peptide binding energy calculations based on structural ensemble and multiple peptide templates

SH3 domains mediate signal transduction by recognizing short peptides. Understanding of the driving forces in peptide recognitions will help us to predict the binding specificity of the domain-peptide recognition and to understand the molecular interaction networks of cells. However, accurate calculation of the binding energy is a tough challenge. In this study, we propose three ideas for improving our ability to predict the binding energy between SH3 domains and peptides: (1) utilizing the structural ensembles sampled from a molecular dynamics simulation trajectory, (2) utilizing multiple peptide templates, and (3) optimizing the sequence-structure mapping. We tested these three ideas on ten previously studied SH3 domains for which SPOT analysis data were available. The results indicate that calculating binding energy using the structural ensemble was most effective, clearly increasing the prediction accuracy, while the second and third ideas tended to give better binding energy predictions. We applied our method to the five SH3 targets in DREAM4 Challenge and selected the best performing method.

Microexon-based regulation of ITSN1 and Src SH3 domains specificity relies on introduction of charged amino acids into the interaction interface

SH3 domains function as protein-protein interaction modules in assembly of signalling and endocytic protein complexes. Here we report investigations of the mechanism of regulation of the binding properties of the SH3 domains of intersectin (ITSN1) and Src kinase by alternative splicing. Comparative sequence analysis of ITSN1 and Src genes revealed the conservation of alternatively spliced microexons affecting the structure of the SH3 domains in vertebrates. We show that neuron-specific ITSN1 transcripts containing the microexon 20 that encodes five amino acid residues within the SH3A domain are expressed in zebrafish from the earliest stages of the development of the nervous system. Models of alternative isoforms of the ITSN1 SH3A domain revealed that the insertion encoded by the microexon is located at the beginning of the n-Src loop of this domain causing a shift of negatively charged amino acids towards the interaction interface. Mutational analysis confirmed the importance of translocation of these negatively charged amino acids for interaction with dynamin 1. We also identified a residue within the microexon-encoded insert in the SH3 domain of brain-specific variant of Src that abolishes interaction of the domain with dynamin 1. Thus microexons provide a mechanism for the control of tissue-specific interactions of ITSN1 and Src with their partners.

The evolution of protein interaction networks in regulatory proteins

Interactions between proteins are essential for intracellular communication. They form complex networks which have become an important source for functional analysis of proteins. Combining phylogenies with network analysis, we investigate the evolutionary history of interaction networks from the bHLH, NR and bZIP transcription-factor families. The bHLH and NR networks show a hub-like structure with varying γ values. Mutation and gene duplication play an important role in adding and removing interactions. We conclude that in several of the protein families that we have studied, networks have primarily arisen by the development of heterodimerizing transcription factors, from an ancestral gene which interacts with any of the newly emerging proteins but also homodimerizes.

iSPOT: a web tool for the analysis and recognition of protein domain specificity

Methods that aim at predicting interaction partners are very likely to play an important role in the interpretation of genomic information. iSPOT (iSpecificity Prediction Of Target) is a web tool (accessible at http://cbm.bio.uniroma2.it/iSPOT) developed for the prediction of protein-protein interaction mediated by families of peptide recognition modules. iSPOT accesses a database of position specific residue-residue interaction frequencies for members of the SH3 and PDZ protein domain families. The software utilises this database to provide a score for any potential domain peptide interaction.

iSPOT: 1. evaluates the likelihood of the interaction between any of the peptides contained in an input protein and a list of domains of the two different families; 2. searches in the SWISS-PROT database for potential partners of a query domain; and 3. has access to a repository of all the domain/target peptide interaction data.

Novel peptide-mediated interactions derived from high-resolution 3-dimensional structures

Many biological responses to intra- and extracellular stimuli are regulated through complex networks of transient protein interactions where a globular domain in one protein recognizes a linear peptide from another, creating a relatively small contact interface. These peptide stretches are often found in unstructured regions of proteins, and contain a consensus motif complementary to the interaction surface displayed by their binding partners. While most current methods for the de novo discovery of such motifs exploit their tendency to occur in disordered regions, our work here focuses on another observation: upon binding to their partner domain, motifs adopt a well-defined structure. Indeed, through the analysis of all peptide-mediated interactions of known high-resolution three-dimensional (3D) structure, we found that the structure of the peptide may be as characteristic as the consensus motif, and help identify target peptides even though they do not match the established patterns. Our analyses of the structural features of known motifs reveal that they tend to have a particular stretched and elongated structure, unlike most other peptides of the same length. Accordingly, we have implemented a strategy based on a Support Vector Machine that uses this features, along with other structure-encoded information about binding interfaces, to search the set of protein interactions of known 3D structure and to identify unnoticed peptide-mediated interactions among them. We have also derived consensus patterns for these interactions, whenever enough information was available, and compared our results with established linear motif patterns and their binding domains. Finally, to cross-validate our identification strategy, we scanned interactome networks from four model organisms with our newly derived patterns to see if any of them occurred more often than expected. Indeed, we found significant over-representations for 64 domain-motif interactions, 46 of which had not been described before, involving over 6,000 interactions in total for which we could suggest the molecular details determining the binding.

Protein-protein interactions are paramount in any aspect of the cellular life. Some proteins form large macromolecular complexes that execute core functionalities of the cell, while others transmit information in signalling networks to co-ordinate these processes. The latter type, of more transient nature, often occurs through the recognition of a small linear sequence motif in one protein by a specialized globular domain in the other. These peptide stretches often contain a consensus pattern complementary to the interaction surface displayed by their binding partners, and adopt a well-defined structure upon binding. Information that is currently available only from high-resolution three-dimensional (3D) structures, and that can be as characteristic as the consensus motif itself. In this manuscript, we present a strategy to identify novel domain-motif interactions (DMIs) among the set of protein complexes of known 3D structures, which provides information on the consensus motif and binding domain and also allows ready identification of the key interacting residues. A detailed knowledge of the interface is critical to plan further functional studies and for the development of interfering elements, be it drug-like compounds or novel engineered binding proteins or peptides. The small interfaces typical for DMIs make them interesting candidates for all these applications.

Modular architecture of nucleotide-binding pockets

Recently, modularity has emerged as a general attribute of complex biological systems. This is probably because modular systems lend themselves readily to optimization via random mutation followed by natural selection. Although they are not traditionally considered to evolve by this process, biological ligands are also modular, being composed of recurring chemical fragments, and moreover they exhibit similarities reminiscent of mutations (e.g. the few atoms differentiating adenine and guanine). Many ligands are also promiscuous in the sense that they bind to many different protein folds. Here, we investigated whether ligand chemical modularity is reflected in an underlying modularity of binding sites across unrelated proteins. We chose nucleotides as paradigmatic ligands, because they can be described as composed of well-defined fragments (nucleobase, ribose and phosphates) and are quite abundant both in nature and in protein structure databases. We found that nucleotide-binding sites do indeed show a modular organization and are composed of fragment-specific protein structural motifs, which parallel the modular structure of their ligands. Through an analysis of the distribution of these motifs in different proteins and in different folds, we discuss the evolutionary implications of these findings and argue that the structural features we observed can arise both as a result of divergence from a common ancestor or convergent evolution.

Accurate prediction of DnaK-peptide binding via homology modelling and experimental data

Molecular chaperones are essential elements of the protein quality control machinery that governs translocation and folding of nascent polypeptides, refolding and degradation of misfolded proteins, and activation of a wide range of client proteins. The prokaryotic heat-shock protein DnaK is the E. coli representative of the ubiquitous Hsp70 family, which specializes in the binding of exposed hydrophobic regions in unfolded polypeptides. Accurate prediction of DnaK binding sites in E. coli proteins is an essential prerequisite to understand the precise function of this chaperone and the properties of its substrate proteins. In order to map DnaK binding sites in protein sequences, we have developed an algorithm that combines sequence information from peptide binding experiments and structural parameters from homology modelling. We show that this combination significantly outperforms either single approach. The final predictor had a Matthews correlation coefficient (MCC) of 0.819 when assessed over the 144 tested peptide sequences to detect true positives and true negatives. To test the robustness of the learning set, we have conducted a simulated cross-validation, where we omit sequences from the learning sets and calculate the rate of repredicting them. This resulted in a surprisingly good MCC of 0.703. The algorithm was also able to perform equally well on a blind test set of binders and non-binders, of which there was no prior knowledge in the learning sets. The algorithm is freely available at http://limbo.vib.be.

Molecular chaperones are essential elements of the protein quality control machinery that governs translocation and folding of nascent polypeptides, refolding and degradation of misfolded proteins, and activation of a wide range of client proteins. This variety of functions results from the existence of multiple chaperones with different structures. Chaperones bind to exposed regions of proteins to fulfil their function. The chaperone must hereby recognise a certain signal sequence on the substrate protein. The nature of the sequence that is exposed will determine the types of chaperones that can interact with it, and in the end will also determine the fate of the substrate protein: refolding, translocation, degradation or activation. Knowledge of the binding sequence determinants of molecular chaperones will shed more light on the mechanism of how each chaperone contributes to the cellular protein quality control system.

In this study we have made an algorithm which accurately predicts binding sites for the well studied E. coli Hsp70 chaperone, DnaK, which is implicated in folding efficiency and prevention of aggregation. The ability to detect and design high-affinity DnaK binding sites enhances our understanding of chaperone-substrate recognition and opens great opportunities to enhance protein solubility using protein-DnaK binding motif fusions.

SH3 domains of Grb2 adaptor bind to PXpsiPXR motifs within the Sos1 nucleotide exchange factor in a discriminate manner

Ubiquitously encountered in a wide variety of cellular processes, the Grb2-Sos1 interaction is mediated through the combinatorial binding of nSH3 and cSH3 domains of Grb2 to various sites containing PXpsiPXR motifs within Sos1. Here, using isothermal titration calorimetry, we demonstrate that while the nSH3 domain binds with affinities in the physiological range to all four sites containing PXpsiPXR motifs, designated S1, S2, S3, and S4, the cSH3 domain can only do so at the S1 site. Further scrutiny of these sites yields rationale for the recognition of various PXpsiPXR motifs by the SH3 domains in a discriminate manner. Unlike the PXpsiPXR motifs at S2, S3, and S4 sites, the PXpsiPXR motif at the S1 site is flanked at its C-terminus with two additional arginine residues that are absolutely required for high-affinity binding of the cSH3 domain. In striking contrast, these two additional arginine residues augment the binding of the nSH3 domain to the S1 site, but their role is not critical for the recognition of S2, S3, and S4 sites. Site-directed mutagenesis suggests that the two additional arginine residues flanking the PXpsiPXR motif at the S1 site contribute to free energy of binding via the formation of salt bridges with specific acidic residues in SH3 domains. Molecular modeling is employed to project these novel findings into the 3D structures of SH3 domains in complex with a peptide containing the PXpsiPXR motif and flanking arginine residues at the S1 site. Taken together, this study furthers our understanding of the assembly of a key signaling complex central to cellular machinery.

Using genome-wide measurements for computational prediction of SH2-peptide interactions

Peptide-recognition modules (PRMs) are used throughout biology to mediate protein–protein interactions, and many PRMs are members of large protein domain families. Recent genome-wide measurements describe networks of peptide–PRM interactions. In these networks, very similar PRMs recognize distinct sets of peptides, raising the question of how peptide-recognition specificity is achieved using similar protein domains. The analysis of individual protein complex structures often gives answers that are not easily applicable to other members of the same PRM family. Bioinformatics-based approaches, one the other hand, may be difficult to interpret physically. Here we integrate structural information with a large, quantitative data set of SH2 domain–peptide interactions to study the physical origin of domain–peptide specificity. We develop an energy model, inspired by protein folding, based on interactions between the amino-acid positions in the domain and peptide. We use this model to successfully predict which SH2 domains and peptides interact and uncover the positions in each that are important for specificity. The energy model is general enough that it can be applied to other members of the SH2 family or to new peptides, and the cross-validation results suggest that these energy calculations will be useful for predicting binding interactions. It can also be adapted to study other PRM families, predict optimal peptides for a given SH2 domain, or study other biological interactions, e.g. protein–DNA interactions.

Domain Interaction Footprint: a multi-classification approach to predict domain-peptide interactions

MOTIVATION

The flow of information within cellular pathways largely relies on specific protein-protein interactions. Discovering such interactions that are mostly mediated by peptide recognition modules (PRM) is therefore a fundamental step towards unravelling the complexity of varying pathways. Since peptides can be recognized by more than one PRM and high-throughput experiments are both time consuming and expensive, it would be preferable to narrow down all potential peptide ligands for one specific PRM by a computational method. We at first present Domain Interaction Footprint (DIF) a new approach to predict binding peptides to PRMs merely based on the sequence of the peptides. Second, we show that our method is able to create a multi-classification model that assesses the binding specificity of a given peptide to all examined PRMs at once.

RESULTS

We first applied our approach to a previously investigated dataset of different SH3 domains and predicted their appropriate peptide ligands with an exceptionally high accuracy. This result outperforms all recent methods trained on the same dataset. Furthermore, we used our technique to build two multi-classification models (SH3 and PDZ domains) to predict the interaction preference between a peptide and every single domain in the corresponding domain family at once. Predicting the domain specificity most reliably, our proposed approach can be seen as a first step towards a complete multi-domain classification model comprised of all domains of one family. Such a comprehensive domain specificity model would benefit the quest for highly specific peptide ligands interacting solely with the domain of choice.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Accurate prediction of peptide binding sites on protein surfaces

Many important protein-protein interactions are mediated by the binding of a short peptide stretch in one protein to a large globular segment in another. Recent efforts have provided hundreds of examples of new peptides binding to proteins for which a three-dimensional structure is available (either known experimentally or readily modeled) but where no structure of the protein-peptide complex is known. To address this gap, we present an approach that can accurately predict peptide binding sites on protein surfaces. For peptides known to bind a particular protein, the method predicts binding sites with great accuracy, and the specificity of the approach means that it can also be used to predict whether or not a putative or predicted peptide partner will bind. We used known protein-peptide complexes to derive preferences, in the form of spatial position specific scoring matrices, which describe the binding-site environment in globular proteins for each type of amino acid in bound peptides. We then scan the surface of a putative binding protein for sites for each of the amino acids present in a peptide partner and search for combinations of high-scoring amino acid sites that satisfy constraints deduced from the peptide sequence. The method performed well in a benchmark and largely agreed with experimental data mapping binding sites for several recently discovered interactions mediated by peptides, including RG-rich proteins with SMN domains, Epstein-Barr virus LMP1 with TRADD domains, DBC1 with Sir2, and the Ago hook with Argonaute PIWI domain. The method, and associated statistics, is an excellent tool for predicting and studying binding sites for newly discovered peptides mediating critical events in biology.

Identification of the variant Ala335Val of MED25 as responsible for CMT2B2: molecular data, functional studies of the SH3 recognition motif and correlation between wild-type MED25 and PMP22 RNA levels in CMT1A animal models

Charcot-Marie-Tooth (CMT) disease is a clinically and genetically heterogeneous disorder. All mendelian patterns of inheritance have been described. We identified a homozygous p.A335V mutation in the MED25 gene in an extended Costa Rican family with autosomal recessively inherited Charcot-Marie-Tooth neuropathy linked to the CMT2B2 locus in chromosome 19q13.3. MED25, also known as ARC92 and ACID1, is a subunit of the human activator-recruited cofactor (ARC), a family of large transcriptional coactivator complexes related to the yeast Mediator. MED25 was identified by virtue of functional association with the activator domains of multiple cellular and viral transcriptional activators. Its exact physiological function in transcriptional regulation remains obscure. The CMT2B2-associated missense amino acid substitution p.A335V is located in a proline-rich region with high affinity for SH3 domains of the Abelson type. The mutation causes a decrease in binding specificity leading to the recognition of a broader range of SH3 domain proteins. Furthermore, Med25 is coordinately expressed with Pmp22 gene dosage and expression in transgenic mice and rats. These results suggest a potential role of this protein in the molecular etiology of CMT2B2 and suggest a potential, more general role of MED25 in gene dosage sensitive peripheral neuropathy pathogenesis.

Characterization of domain-peptide interaction interface: a generic structure-based model to decipher the binding specificity of SH3 domains

Extensive efforts have been devoted to determining the binding specificity of Src homology 3 (SH3) domains usually in a case-by-case manner. A generic structure-based model is necessary to decipher the protein recognition code of the entire domain family. In this study, we have developed a general framework that combines molecular modeling and a machine learning algorithm to capture the energetic characteristics of the domain-peptide interactions and predict the binding specificity of the SH3 domain family. Our model is not trained for individual SH3 domains; rather it is a generic model for the entire domain family. Our model not only achieved satisfactory prediction accuracy but also provided structural insights into which residues are important for the binding specificity. The success of our framework on SH3 domains suggests that it is possible to establish a theoretical model to decipher the protein recognition code of any modular domain.

Predicting helix orientation for coiled-coil dimers

The alpha-helical coiled coil is a structurally simple protein oligomerization or interaction motif consisting of two or more alpha helices twisted into a supercoiled bundle. Coiled coils can differ in their stoichiometry, helix orientation, and axial alignment. Because of the near degeneracy of many of these variants, coiled coils pose a challenge to fold recognition methods for structure prediction. Whereas distinctions between some protein folds can be discriminated on the basis of hydrophobic/polar patterning or secondary structure propensities, the sequence differences that encode important details of coiled-coil structure can be subtle. This is emblematic of a larger problem in the field of protein structure and interaction prediction: that of establishing specificity between closely similar structures. We tested the behavior of different computational models on the problem of recognizing the correct orientation--parallel vs. antiparallel--of pairs of alpha helices that can form a dimeric coiled coil. For each of 131 examples of known structure, we constructed a large number of both parallel and antiparallel structural models and used these to assess the ability of five energy functions to recognize the correct fold. We also developed and tested three sequence-based approaches that make use of varying degrees of implicit structural information. The best structural methods performed similarly to the best sequence methods, correctly categorizing approximately 81% of dimers. Steric compatibility with the fold was important for some coiled coils we investigated. For many examples, the correct orientation was determined by smaller energy differences between parallel and antiparallel structures distributed over many residues and energy components. Prediction methods that used structure but incorporated varying approximations and assumptions showed quite different behaviors when used to investigate energetic contributions to orientation preference. Sequence based methods were sensitive to the choice of residue-pair interactions scored.

Identification and rational redesign of peptide ligands to CRIP1, a novel biomarker for cancers

Cysteine-rich intestinal protein 1 (CRIP1) has been identified as a novel marker for early detection of cancers. Here we report on the use of phage display in combination with molecular modeling to identify a high-affinity ligand for CRIP1. Panning experiments using a circularized C7C phage library yielded several consensus sequences with modest binding affinities to purified CRIP1. Two sequence motifs, A1 and B5, having the highest affinities for CRIP1, were chosen for further study. With peptide structure information and the NMR structure of CRIP1, the higher-affinity A1 peptide was computationally redesigned, yielding a novel peptide, A1M, whose affinity was predicted to be much improved. Synthesis of the peptide and saturation and competitive binding studies demonstrated approximately a 10–28-fold improvement in the affinity of A1M compared to that of either A1 or B5 peptide. These techniques have broad application to the design of novel ligand peptides.

Breast cancer is one of the most frequently diagnosed malignancies in American females and is the second leading cause of cancer deaths in women. Several improvements in diagnostic protocols have enhanced our ability for earlier detection of breast cancer, resulting in improvement of therapeutic outcome and an increased survival rate for breast cancer patients. However, current early screening techniques are neither comprehensive nor infallible. Imaging techniques that improve breast cancer detection, localization, and evaluation of therapy are essential in combating the disease. Cysteine-rich intestinal protein 1 (CRIP1) has been identified as a novel marker for early detection of breast cancers. Here, we report the use of phage display and computational molecular modeling to identify a high-affinity ligand for CRIP1. Phage display panning experiments initially identified consensus peptide sequences with modest binding affinity to purified CRIP1. Using ab initio modeling of binding peptide structures, computational docking, and recently developed free energy estimation protocols, we redesigned the peptides to increase their affinity for CRIP1. Synthesis of the redesigned peptide and binding studies demonstrated approximately a 10–28-fold improvement in the binding affinity. The combination of computational and experimental techniques in this study demonstrates a potentially powerful tool in modulating protein–protein interactions.

Peptide-mediated interactions in biological systems: new discoveries and applications

Peptide-mediated interactions play very important roles in cellular processes. Recent years have seen much activity in the discovery of new bioactive peptides, and interactions mediated by protein-peptide binding events. At the same time, computational approaches continue to be developed that allow protein-peptide interactions to be discovered with great accuracy. There are also a growing number of chemicals that can target these interactions with various applications in disease. Both new discoveries and predictions suggest that these protein-peptide interactions play greater roles in cellular processes than previously thought. We propose that projects to uncover the protein-peptide repertoire used in Nature in a systematic way will have numerous applications in molecular biology and medicine.

Cytoskeleton assembly at endothelial cell-cell contacts is regulated by alphaII-spectrin-VASP complexes

Directed cortical actin assembly is the driving force for intercellular adhesion. Regulated by phosphorylation, vasodilator-stimulated phosphoprotein (VASP) participates in actin fiber formation. We screened for endothelial proteins, which bind to VASP, dependent on its phosphorylation status. Differential proteomics identified αII-spectrin as such a VASP-interacting protein. αII-Spectrin binds to the VASP triple GP₅-motif via its SH3 domain. cAMP-dependent protein kinase–mediated VASP phosphorylation at Ser157 inhibits αII-spectrin–VASP binding. VASP is dephosphorylated upon formation of cell–cell contacts and in confluent, but not in sparse cells, αII-spectrin colocalizes with nonphosphorylated VASP at cell–cell junctions. Ectopic expression of the αII-spectrin SH3 domain at cell–cell contacts translocates VASP, initiates cortical actin cytoskeleton formation, stabilizes cell–cell contacts, and decreases endothelial permeability. Conversely, the permeability of VASP-deficient endothelial cells (ECs) and microvessels of VASP-null mice increases. Reconstitution of VASP-deficient ECs rescues barrier function, whereas αII-spectrin binding-deficient VASP mutants fail to restore elevated permeability. We propose that αII-spectrin–VASP complexes regulate cortical actin cytoskeleton assembly with implications for vascular permeability.

SH3-Hunter: discovery of SH3 domain interaction sites in proteins

SH3-Hunter (http://cbm.bio.uniroma2.it/SH3-Hunter/) is a web server for the recognition of putative SH3 domain interaction sites on protein sequences. Given an input query consisting of one or more protein sequences, the server identifies peptides containing poly-proline binding motifs and associates them to a list of SH3 domains, in order to compose peptide-domain pairs. The server can accept a list of peptides and allows users to upload an input file in a proper format. An accurate selection of SH3 domains is available and users can also submit their own SH3 domain sequence. SH3-Hunter evaluates which peptide-domain pair represents a possible interaction pair and produces as output a list of significant interaction sites for each query protein. Each proposed interaction site is associated to a propensity score and sensitivity and precision levels for the prediction. The server prediction capability is based on a neural network model integrating high-throughput pep-spot data with structural information extracted from known SH3-peptide complexes.

Nonnatural protein-protein interaction-pair design by key residues grafting

Protein-protein interface design is one of the most exciting fields in protein science; however, designing nonnatural protein-protein interaction pairs remains difficult. In this article we report a de novo design of a nonnatural protein-protein interaction pair by scanning the Protein Data Bank for suitable scaffold proteins that can be used for grafting key interaction residues and can form stable complexes with the target protein after additional mutations. Using our design algorithm, an unrelated protein, rat PLCdelta(1)-PH (pleckstrin homology domain of phospholipase C-delta1), was successfully designed to bind the human erythropoietin receptor (EPOR) after grafting the key interaction residues of human erythropoietin binding to EPOR. The designed mutants of rat PLCdelta(1)-PH were expressed and purified to test their binding affinities with EPOR. A designed triple mutation of PLCdelta(1)-PH (ERPH1) was found to bind EPOR with high affinity (K(D) of 24 nM and an IC(50) of 5.7 microM) both in vitro and in a cell-based assay, respectively, although the WT PLCdelta(1)-PH did not show any detectable binding under the assay conditions. The in vitro binding affinities of the PLCdelta(1)-PH mutants correlate qualitatively to the computational binding affinities, validating the design and the protein-protein interaction model. The successful practice of finding a proper protein scaffold and making it bind with EPOR demonstrates a prospective application in protein engineering targeting protein-protein interfaces.

Solution structure of a Hck SH3 domain ligand complex reveals novel interaction modes

We studied the interaction of hematopoietic cell kinase SH3 domain (HckSH3) with an artificial 12-residue proline-rich peptide PD1 (HSKYPLPPLPSL) identified as high affinity ligand (K(D)=0.2 muM). PD1 shows an unusual ligand sequence for SH3 binding in type I orientation because it lacks the typical basic anchor residue at position P(-3), but instead has a tyrosine residue at this position. A basic lysine residue, however, is present at position P(-4). The solution structure of the HckSH3:PD1 complex, which is the first HckSH3 complex structure available, clearly reveals that the P(-3) tyrosine residue of PD1 does not take the position of the typical anchor residue but rather forms additional van der Waals interactions with the HckSH3 RT loop. Instead, lysine at position P(-4) of PD1 substitutes the function of the P(-3) anchor residue. This finding expands the well known ligand consensus sequence +xxPpxP by +xxxPpxP. Thus, software tools like iSPOT fail to identify PD1 as a high-affinity HckSH3 ligand so far. In addition, a short antiparallel beta-sheet in the RT loop of HckSH3 is observed upon PD1 binding. The structure of the HckSH3:PD1 complex reveals novel features of SH3 ligand binding and yields new insights into the structural basics of SH3-ligand interactions. Consequences for computational prediction tools adressing SH3-ligand interactions as well as the biological relevance of our findings are discussed.

Computational Design of a New Hydrogen Bond Network and at Least a 300-fold Specificity Switch at a Protein−Protein Interface

The redesign of protein-protein interactions is a stringent test of our understanding of molecular recognition and specificity. Previously we engineered a modest specificity switch into the colicin E7 DNase-Im7 immunity protein complex by identifying mutations that are disruptive in the native complex, but can be compensated by mutations on the interacting partner. Here we extend the approach by systematically sampling alternate rigid body orientations to optimize the interactions in a binding mode specific manner. Using this protocol we designed a de novo hydrogen bond network at the DNase-immunity protein interface and confirmed the design with X-ray crystallographic analysis. Subsequent design of the second shell of interactions guided by insights from the crystal structure on tightly bound water molecules, conformational strain, and packing defects yielded new binding partners that exhibited specificities of at least 300-fold between the cognate and the non-cognate complexes. This multi-step approach should be applicable to the design of polar protein-protein interactions and contribute to the re-engineering of regulatory networks mediated by protein-protein interactions.

A novel structure-based encoding for machine-learning applied to the inference of SH3 domain specificity

MOTIVATION

Unravelling the rules underlying protein-protein and protein-ligand interactions is a crucial step in understanding cell machinery. Peptide recognition modules (PRMs) are globular protein domains which focus their binding targets on short protein sequences and play a key role in the frame of protein-protein interactions. High-throughput techniques permit the whole proteome scanning of each domain, but they are characterized by a high incidence of false positives. In this context, there is a pressing need for the development of in silico experiments to validate experimental results and of computational tools for the inference of domain-peptide interactions.

RESULTS

We focused on the SH3 domain family and developed a machine-learning approach for inferring interaction specificity. SH3 domains are well-studied PRMs which typically bind proline-rich short sequences characterized by the PxxP consensus. The binding information is known to be held in the conformation of the domain surface and in the short sequence of the peptide. Our method relies on interaction data from high-throughput techniques and benefits from the integration of sequence and structure data of the interacting partners. Here, we propose a novel encoding technique aimed at representing binding information on the basis of the domain-peptide contact residues in complexes of known structure. Remarkably, the new encoding requires few variables to represent an interaction, thus avoiding the 'curse of dimension'. Our results display an accuracy >90% in detecting new binders of known SH3 domains, thus outperforming neural models on standard binary encodings, profile methods and recent statistical predictors. The method, moreover, shows a generalization capability, inferring specificity of unknown SH3 domains displaying some degree of similarity with the known data.

Identification of preferred protein interactions by phage-display of the human Src homology-3 proteome

We have determined the human genome to contain 296 different Src homology-3 (SH3) domains and cloned them into a phage-display vector. This provided a powerful and unbiased system for simultaneous assaying of the complete human SH3 proteome for the strongest binding to target proteins of interest, without the limitations posed by short linear peptide ligands or confounding variables of more indirect methods for protein interaction screening. Studies involving three ligand proteins, human immunodeficiency virus-1 Nef, p21-activated kinase (PAK)2 and ADAM15, showed previously reported as well as novel SH3 partners with nanomolar affinities specific for them. This argues that SH3 domains may have a more dominant role in directing cellular protein interactions than has been assumed. Besides showing potentially important new SH3-directed interactions, these studies also led to the discovery of novel signalling proteins, such as the PAK2-binding adaptor protein POSH2 and the ADAM15-binding sorting nexin family member SNX30.

An integrated machine learning system to computationally screen protein databases for protein binding peptide ligands

A fairly large set of protein interactions is mediated by families of peptide binding domains, such as Src homology 2 (SH2), SH3, PDZ, major histocompatibility complex, etc. To identify their ligands by experimental screening is not only labor-intensive but almost futile in screening low abundance species due to the suppression by high abundance species. An ideal way of studying protein-protein interactions is to use high throughput computational approaches to screen protein sequence databases to direct the validating experiments toward the most promising peptides. Predictors with only good cross-validation were not good enough to screen protein databases. In the current study we built integrated machine learning systems using three novel coding methods and screened the Swiss-Prot and GenBank protein databases for potential ligands of 10 SH3 and three PDZ domains. A large fraction of predictions has already been experimentally confirmed by other independent research groups, indicating a satisfying generalization capability for future applications in identifying protein interactions.

Structure-based Prediction of bZIP Partnering Specificity

Predicting protein interaction specificity from sequence is an important goal in computational biology. We present a model for predicting the interaction preferences of coiled-coil peptides derived from bZIP transcription factors that performs very well when tested against experimental protein microarray data. We used only sequence information to build atomic-resolution structures for 1711 dimeric complexes, and evaluated these with a variety of functions based on physics, learned empirical weights or experimental coupling energies. A purely physical model, similar to those used for protein design studies, gave reasonable performance. The results were improved significantly when helix propensities were used in place of a structurally explicit model to represent the unfolded reference state. Further improvement resulted upon accounting for residue-residue interactions in competing states in a generic way. Purely physical structure-based methods had difficulty capturing core interactions accurately, especially those involving polar residues such as asparagine. When these terms were replaced with weights from a machine-learning approach, the resulting model was able to correctly order the stabilities of over 6000 pairs of complexes with greater than 90% accuracy. The final model is physically interpretable, and suggests specific pairs of residues that are important for bZIP interaction specificity. Our results illustrate the power and potential of structural modeling as a method for predicting protein interactions and highlight obstacles that must be overcome to reach quantitative accuracy using a de novo approach. Our method shows unprecedented performance in predicting protein-protein interaction specificity accurately using structural modeling and suggests that predicting coiled-coil interactions generally may be within reach.

Computational analysis and prediction of the binding motif and protein interacting partners of the Abl SH3 domain

Protein-protein interactions, particularly weak and transient ones, are often mediated by peptide recognition domains, such as Src Homology 2 and 3 (SH2 and SH3) domains, which bind to specific sequence and structural motifs. It is important but challenging to determine the binding specificity of these domains accurately and to predict their physiological interacting partners. In this study, the interactions between 35 peptide ligands (15 binders and 20 non-binders) and the Abl SH3 domain were analyzed using molecular dynamics simulation and the Molecular Mechanics/Poisson-Boltzmann Solvent Area method. The calculated binding free energies correlated well with the rank order of the binding peptides and clearly distinguished binders from non-binders. Free energy component analysis revealed that the van der Waals interactions dictate the binding strength of peptides, whereas the binding specificity is determined by the electrostatic interaction and the polar contribution of desolvation. The binding motif of the Abl SH3 domain was then determined by a virtual mutagenesis method, which mutates the residue at each position of the template peptide relative to all other 19 amino acids and calculates the binding free energy difference between the template and the mutated peptides using the Molecular Mechanics/Poisson-Boltzmann Solvent Area method. A single position mutation free energy profile was thus established and used as a scoring matrix to search peptides recognized by the Abl SH3 domain in the human genome. Our approach successfully picked ten out of 13 experimentally determined binding partners of the Abl SH3 domain among the top 600 candidates from the 218,540 decapeptides with the PXXP motif in the SWISS-PROT database. We expect that this physical-principle based method can be applied to other protein domains as well.

One of the central questions of molecular biology is to understand how signals are transduced in the cell. Intracellular signal transduction is mainly achieved through cascades of protein-protein interactions, which are often mediated by peptide-binding modular domains, such as Src Homology 2 and 3 (SH2 and SH3). Each family of these domains binds to peptides with specific sequence and structural characteristics. To reconstruct the protein-protein interaction networks mediated by modular domains, one must identify the peptide motifs recognized by these domains and understand the mechanism of binding specificity. These questions are challenging because the domain-peptide interactions are usually weak and transient. Here, the authors took a physical-principles approach to address these difficult questions for the SH3 domain of human protein Abl, which binds to peptides containing the PXXP motif (where P is proline and X is any amino acid). They generated a position-specific scoring matrix to represent the binding motif of the Abl SH3 domain. Analysis on the binding free energy components suggested insights into how the binding specificity is achieved. Most known protein interacting partners of the Abl SH3 domain were correctly identified using the position-specific scoring matrix, and other potential interacting partners were also suggested.

Insights into human Lck SH3 domain binding specificity: different binding modes of artificial and native ligands

We analyzed the ligand binding specificity of the lymphocyte specific kinase (Lck) SH3 domain. We identified artificial Lck SH3 ligands using phage display. In addition, we analyzed Lck SH3 binding sites within known natural Lck SH3 binding proteins using an Lck specific binding assay on membrane-immobilized synthetic peptides. On one hand, from the phage-selected peptides, representing mostly special class I' ligands, a well-defined consensus sequence was obtained. Interestingly, a histidine outside the central polyproline motif contributes significantly to Lck SH3 binding affinity and specificity. On the other hand, we confirmed previously mapped Lck SH3 binding sites in ADAM15, HS1, SLP76, and NS5A, and identified putative Lck SH3 binding sites of Sam68, FasL, c-Cbl, and Cbl-b. Without exception, the comparatively diverse Lck SH3 binding sites of all analyzed natural Lck SH3 binding proteins emerged as class II proteins. Possible explanations for the observed variations between artificial and native ligands-which are not due to significant K(D) value differences as shown by calculating Lck SH3 affinities of artificial peptide PD1-Y(-3)R as well as for peptides comprising putative Lck SH3 binding sites of NS5A, Sos, and Sam68-are discussed. Our data suggest that phage display, a popular tool for determining SH3 binding specificity, must-at least in the case of Lck-not irrevocably mirror physiologically relevant protein-ligand interactions.

Substrate specificity of protein kinases and computational prediction of substrates

To ensure signalling fidelity, kinases must act only on a defined subset of cellular targets. Appreciating the basis for this substrate specificity is essential for understanding the role of an individual protein kinase in a particular cellular process. The specificity in the cell is determined by a combination of "peptide specificity" of the kinase (the molecular recognition of the sequence surrounding the phosphorylation site), substrate recruitment and phosphatase activity. Peptide specificity plays a crucial role and depends on the complementarity between the kinase and the substrate and therefore on their three-dimensional structures. Methods for experimental identification of kinase substrates and characterization of specificity are expensive and laborious, therefore, computational approaches are being developed to reduce the amount of experimental work required in substrate identification. We discuss the structural basis of substrate specificity of protein kinases and review the experimental and computational methods used to obtain specificity information.

A neural strategy for the inference of SH3 domain-peptide interaction specificity

Background

The SH3 domain family is one of the most representative and widely studied cases of so-called Peptide Recognition Modules (PRM). The polyproline II motif PxxP that generally characterizes its ligands does not reflect the complex interaction spectrum of the over 1500 different SH3 domains, and the requirement of a more refined knowledge of their specificity implies the setting up of appropriate experimental and theoretical strategies. Due to the limitations of the current technology for peptide synthesis, several experimental high-throughput approaches have been devised to elucidate protein-protein interaction mechanisms. Such approaches can rely on and take advantage of computational techniques, such as regular expressions or position specific scoring matrices (PSSMs) to pre-process entire proteomes in the search for putative SH3 targets.

In this regard, a reliable inference methodology to be used for reducing the sequence space of putative binding peptides represents a valuable support for molecular and cellular biologists.

Results

Using as benchmark the peptide sequences obtained from in vitro binding experiments, we set up a neural network model that performs better than PSSM in the detection of SH3 domain interactors. In particular our model is more precise in its predictions, even if its performance can vary among different SH3 domains and is strongly dependent on the number of binding peptides in the benchmark.

Conclusion

We show that a neural network can be more effective than standard methods in SH3 domain specificity detection. Neural classifiers identify general SH3 domain binders and domain-specific interactors from a PxxP peptide population, provided that there are a sufficient proportion of true positives in the training sets. This capability can also improve peptide selection for library definition in array experiments. Further advances can be achieved, including properly encoded domain sequences and structural information as input for a global neural network.