Complex-type-dependent scoring functions in protein-protein docking

A major challenge in the field of protein-protein docking is to discriminate between the many wrong and few near-native conformations, i.e. scoring. Here, we introduce combinatorial complex-type-dependent scoring functions for different types of protein-protein complexes, protease/inhibitor, antibody/antigen, enzyme/inhibitor and others. The scoring functions incorporate both physical and knowledge-based potentials, i.e. atomic contact energy (ACE), the residue pair potential (RP), electrostatic and van der Waals' interactions. For different type complexes, the weights of the scoring functions were optimized by the multiple linear regression method, in which only top 300 structures with ligand root mean square deviation (L_RMSD) less than 20 A from the bound (co-crystallized) docking of 57 complexes were used to construct a training set. We employed the bound docking studies to examine the quality of the scoring function, and also extend to the unbound (separately crystallized) docking studies and extra 8 protein-protein complexes. In bound docking of the 57 cases, the first hits of protease/inhibitor cases are all ranked in the top 5. For the cases of antibody/antigen, enzyme/inhibitor and others, there are 17/19, 5/6 and 13/15 cases with the first hits ranked in the top 10, respectively. In unbound docking studies, the first hits of 9/17 protease/inhibitor, 6/19 antibody/antigen, 1/6 enzyme/inhibitor and 6/15 others' complexes are ranked in the top 10. Additionally, for the extra 8 cases, the first hits of the two protease/inhibitor cases are ranked in the top for the bound and unbound test. For the two enzyme/inhibitor cases, the first hits are ranked 1st for bound test, and the 119th and 17th for the unbound test. For the others, the ranks of the first hits are the 1st for the bound test and the 12th for the 1WQ1 unbound test. To some extent, the results validated our divide-and-conquer strategy in the docking study, which might hopefully shed light on the prediction of protein-protein interactions.     

Combination of scoring schemes for protein docking

Background  

Docking algorithms are developed to predict in which orientation two proteins are likely to bind under natural conditions. The currently used methods usually consist of a sampling step followed by a scoring step. We developed a weighted geometric correlation based on optimised atom specific weighting factors and combined them with our previously published amino acid specific scoring and with a comprehensive SVM-based scoring function.     

An interaction-motif-based scoring function for protein-ligand docking

Background  

A good scoring function is essential for molecular docking computations. In conventional scoring functions, energy terms modeling pairwise interactions are cumulatively summed, and the best docking solution is selected. Here, we propose to transform protein-ligand interactions into three-dimensional geometric networks, from which recurring network substructures, or network motifs, are selected and used to provide probability-ranked interaction templates with which to score docking solutions.     

Evaluating scoring functions for docking and designing beta-secretase inhibitors

Several simple scoring methods were examined for 2 series of beta-secretase (BACE-1) inhibitors to identify a docking/scoring protocol which could be used to design BACE-1 inhibitors in a drug discovery program. Both the PLP1 score and MMFFs interaction energy (E(inter)) performed as well or better than more computationally intensive methods for a set of substrate-based inhibitors, while the latter performed well for both sets of inhibitors.     

Amino acid network and its scoring application in protein-protein docking

Protein-protein complex, composed of hydrophobic and hydrophilic residues, can be divided into hydrophobic and hydrophilic amino acid network structures respectively. In this paper, we are interested in analyzing these two different types of networks and find that these networks are of small-world properties. Due to the characteristic complementarity of the complex interfaces, protein-protein docking can be viewed as a particular network rewiring. These networks of correct docked complex conformations have much more increase of the degree values and decay of the clustering coefficients than those of the incorrect ones. Therefore, two scoring terms based on the network parameters are proposed, in which the geometric complementarity, hydrophobic-hydrophobic and polar-polar interactions are taken into account. Compared with a two-term energy function, a simple scoring function HPNet which includes the two network-based scoring terms shows advantages in two aspects, not relying on energy considerations and better discrimination. Furthermore, combing the network-based scoring terms with some other energy terms, a new multi-term scoring function HPNet-combine can also make some improvements to the scoring function of RosettaDock.     

A collaborative filtering approach for protein-protein docking scoring functions

A protein-protein docking procedure traditionally consists in two successive tasks: a search algorithm generates a large number of candidate conformations mimicking the complex existing in vivo between two proteins, and a scoring function is used to rank them in order to extract a native-like one. We have already shown that using Voronoi constructions and a well chosen set of parameters, an accurate scoring function could be designed and optimized. However to be able to perform large-scale in silico exploration of the interactome, a near-native solution has to be found in the ten best-ranked solutions. This cannot yet be guaranteed by any of the existing scoring functions. In this work, we introduce a new procedure for conformation ranking. We previously developed a set of scoring functions where learning was performed using a genetic algorithm. These functions were used to assign a rank to each possible conformation. We now have a refined rank using different classifiers (decision trees, rules and support vector machines) in a collaborative filtering scheme. The scoring function newly obtained is evaluated using 10 fold cross-validation, and compared to the functions obtained using either genetic algorithms or collaborative filtering taken separately. This new approach was successfully applied to the CAPRI scoring ensembles. We show that for 10 targets out of 12, we are able to find a near-native conformation in the 10 best ranked solutions. Moreover, for 6 of them, the near-native conformation selected is of high accuracy. Finally, we show that this function dramatically enriches the 100 best-ranking conformations in near-native structures.     

Exhaustive sampling of docking poses reveals binding hypotheses for propafenone type inhibitors of P-glycoprotein

Overexpression of the xenotoxin transporter P-glycoprotein (P-gp) represents one major reason for the development of multidrug resistance (MDR), leading to the failure of antibiotic and cancer therapies. Inhibitors of P-gp have thus been advocated as promising candidates for overcoming the problem of MDR. However, due to lack of a high-resolution structure the concrete mode of interaction of both substrates and inhibitors is still not known. Therefore, structure-based design studies have to rely on protein homology models. In order to identify binding hypotheses for propafenone-type P-gp inhibitors, five different propafenone derivatives with known structure-activity relationship (SAR) pattern were docked into homology models of the apo and the nucleotide-bound conformation of the transporter. To circumvent the uncertainty of scoring functions, we exhaustively sampled the pose space and analyzed the poses by combining information retrieved from SAR studies with common scaffold clustering. The results suggest propafenone binding at the transmembrane helices 5, 6, 7 and 8 in both models, with the amino acid residue Y307 playing a crucial role. The identified binding site in the non-energized state is overlapping with, but not identical to, known binding areas of cyclic P-gp inhibitors and verapamil. These findings support the idea of several small binding sites forming one large binding cavity. Furthermore, the binding hypotheses for both catalytic states were analyzed and showed only small differences in their protein-ligand interaction fingerprints, which indicates only small movements of the ligand during the catalytic cycle.     

Prediction of multimolecular assemblies by multiple docking

The majority of proteins function when associated in multimolecular assemblies. Yet, prediction of the structures of multimolecular complexes has largely not been addressed, probably due to the magnitude of the combinatorial complexity of the problem. Docking applications have traditionally been used to predict pairwise interactions between molecules. We have developed an algorithm that extends the application of docking to multimolecular assemblies. We apply it to predict quaternary structures of both oligomers and multi-protein complexes. The algorithm predicted well a near-native arrangement of the input subunits for all cases in our data set, where the number of the subunits of the different target complexes varied from three to ten. In order to simulate a more realistic scenario, unbound cases were tested. In these cases the input conformations of the subunits are either unbound conformations of the subunits or a model obtained by a homology modeling technique. The successful predictions of the unbound cases, where the input conformations of the subunits are different from their conformations within the target complex, suggest that the algorithm is robust. We expect that this type of algorithm should be particularly useful to predict the structures of large macromolecular assemblies, which are difficult to solve by experimental structure determination.     

Prediction of protein-protein interactions by docking methods

Recently, developments have been made in predicting the structure of docked complexes when the coordinates of the components are known. The process generally consists of a stage during which the components are combined rigidly and then a refinement stage. Several rapid new algorithms have been introduced in the rigid docking problem and promising refinement techniques have been developed, based on modified molecular mechanics force fields and empirical measures of desolvation, combined with minimisations that switch on the short-range interactions gradually. There has also been progress in developing a benchmark set of targets for docking and a blind trial, similar to the trials of protein structure prediction, has taken place.     

On docking, scoring and assessing protein-DNA complexes in a rigid-body framework

We consider the identification of interacting protein-nucleic acid partners using the rigid body docking method FTdock, which is systematic and exhaustive in the exploration of docking conformations. The accuracy of rigid body docking methods is tested using known protein-DNA complexes for which the docked and undocked structures are both available. Additional tests with large decoy sets probe the efficacy of two published statistically derived scoring functions that contain a huge number of parameters. In contrast, we demonstrate that state-of-the-art machine learning techniques can enormously reduce the number of parameters required, thereby identifying the relevant docking features using a miniscule fraction of the number of parameters in the prior works. The present machine learning study considers a 300 dimensional vector (dependent on only 15 parameters), termed the Chemical Context Profile (CCP), where each dimension reflects a specific type of protein amino acid-nucleic acid base interaction. The CCP is designed to capture the chemical complementarities of the interface and is well suited for machine learning techniques. Our objective function is the Chemical Context Discrepancy (CCD), which is defined as the angle between the native system's CCP vector and the decoy's vector and which serves as a substitute for the more commonly used root mean squared deviation (RMSD). We demonstrate that the CCP provides a useful scoring function when certain dimensions are properly weighted. Finally, we explore how the amino acids on a protein's surface can help guide DNA binding, first through long-range interactions, followed by direct contacts, according to specific preferences for either the major or minor grooves of the DNA.     

A model binding site for testing scoring functions in molecular docking

Prediction of interaction energies between ligands and their receptors remains a major challenge for structure-based inhibitor discovery. Much effort has been devoted to developing scoring schemes that can successfully rank the affinities of a diverse set of possible ligands to a binding site for which the structure is known. To test these scoring functions, well-characterized experimental systems can be very useful. Here, mutation-created binding sites in T4 lysozyme were used to investigate how the quality of atomic charges and solvation energies affects molecular docking. Atomic charges and solvation energies were calculated for 172,118 molecules in the Available Chemicals Directory using a semi-empirical quantum mechanical approach by the program AMSOL. The database was first screened against the apolar cavity site created by the mutation Leu99Ala (L99A). Compared to the electronegativity-based charges that are widely used, the new charges and desolvation energies improved ranking of known apolar ligands, and better distinguished them from more polar isosteres that are not observed to bind. To investigate whether the new charges had predictive value, the non-polar residue Met102, which forms part of the binding site, was changed to the polar residue glutamine. The structure of the resulting Leu99Ala and Met102Gln double mutant of T4 lysozyme (L99A/M102Q) was determined and the docking calculation was repeated for the new site. Seven representative polar molecules that preferentially docked to the polar versus the apolar binding site were tested experimentally. All seven bind to the polar cavity (L99A/M102Q) but do not detectably bind to the apolar cavity (L99A). Five ligand-bound structures of L99A/M102Q were determined by X-ray crystallography. Docking predictions corresponded to the crystallographic results to within 0.4A RMSD. Improved treatment of partial atomic charges and desolvation energies in database docking appears feasible and leads to better distinction of true ligands. Simple model binding sites, such as L99A and its more polar variants, may find broad use in the development and testing of docking algorithms.     

A new protein-protein docking scoring function based on interface residue properties

MOTIVATION: Protein-protein complexes are known to play key roles in many cellular processes. However, they are often not accessible to experimental study because of their low stability and difficulty to produce the proteins and assemble them in native conformation. Thus, docking algorithms have been developed to provide an in silico approach of the problem. A protein-protein docking procedure traditionally consists of two successive tasks: a search algorithm generates a large number of candidate solutions, and then a scoring function is used to rank them. RESULTS: To address the second step, we developed a scoring function based on a Vorono? tessellation of the protein three-dimensional structure. We showed that the Vorono? representation may be used to describe in a simplified but useful manner, the geometric and physico-chemical complementarities of two molecular surfaces. We measured a set of parameters on native protein-protein complexes and on decoys, and used them as attributes in several statistical learning procedures: a logistic function, Support Vector Machines (SVM), and a genetic algorithm. For the later, we used ROGER, a genetic algorithm designed to optimize the area under the receiver operating characteristics curve. To further test the scores derived with ROGER, we ranked models generated by two different docking algorithms on targets of a blind prediction experiment, improving in almost all cases the rank of native-like solutions. AVAILABILITY: http://genomics.eu.org/spip/-Bioinformatics-tools-     

A filter enhanced sampling and combinatorial scoring study for protein docking in CAPRI

Protein-protein docking is usually exploited with a two-step strategy, i.e., conformational sampling and decoy scoring. In this work, a new filter enhanced sampling scheme was proposed and added into the RosettaDock algorithm to improve the conformational sampling efficiency. The filter term is based on the statistical result that backbone hydrogen bonds in the native protein structures are wrapped by more than nine hydrophobic groups to shield them from attacks of water molecules (Fernandez and Scheraga, Proc Natl Acad Sci USA 2003;100:113-118). A combinatorial scoring function, ComScore, specially designed for the other-type protein-protein complexes was also adopted to select the near native docked modes. ComScore was composed of the atomic contact energy, van der Waals, and electrostatic interaction energies, and the weight of each item was fit through the multiple linear regression approach. To analyze our docking results, the filter enhanced sampling scheme was applied to targets T12, T20, and T21 after the CAPRI blind test, and improvements were obtained. The ligand least root mean square deviations (L_rmsds) were reduced and the hit numbers were increased. ComScore was used in the scoring test for CAPRI rounds 9-12 with good success in rounds 9 and 11.     

Low-resolution docking: Prediction of complexes for underdetermined structures

One of the most fundamental questions concerning ligand-receptor interaction is whether such a process of intermolecular association is generally determined by local structural elements of the participating molecules, or whether there are also large-scale motifs in molecule structures that facilitate complex formation. From the point of view of practical docking computations, the elaborate character of local structural details in ligand-receptor interaction creates a large number of false-positive matches, which interfere with determination of the best fit. Another significant obstacle in protein docking is the problem of structural data inaccuracy (poor structure resolution, conformational changes upon complex formation, etc.). Our study [Vakser (1995) Protein Eng., 8, 371–377], based on ultralow (∼7 Å resolution) representation of molecular structures, allowes to average all high-resolution structural details, and still predict most of the structural features of the ligand-receptor complex. The approach dramatically improves the signal-to-noise ratio in determination of the best fit, and moves the structure inaccuracy tolerance to the range of the macrostructure. In the present paper, we describe a further validation of the main principles of this approach and a detailed analysis of the low-resolution docking results. This includes clustering of ligand positions around the receptor molecule and cross-validation of ligands and receptors from different complexes. We also discuss the important implications of the approach to the multiple-minima problem and a possible role of different structural elements in the recognition mechanism. © 1996 John Wiley & Sons, Inc.     

Prediction of membrane protein types from sequences and position-specific scoring matrices

Membrane protein plays an important role in some biochemical process such as signal transduction, transmembrane transport, etc. Membrane proteins are usually classified into five types [Chou, K.C., Elrod, D.W., 1999. Prediction of membrane protein types and subcellular locations. Proteins: Struct. Funct. Genet. 34, 137-153] or six types [Chou, K.C., Cai, Y.D., 2005. J. Chem. Inf. Modelling 45, 407-413]. Designing in silico methods to identify and classify membrane protein can help us understand the structure and function of unknown proteins. This paper introduces an integrative approach, IAMPC, to classify membrane proteins based on protein sequences and protein profiles. These modules extract the amino acid composition of the whole profiles, the amino acid composition of N-terminal and C-terminal profiles, the amino acid composition of profile segments and the dipeptide composition of the whole profiles. In the computational experiment, the overall accuracy of the proposed approach is comparable with the functional-domain-based method. In addition, the performance of the proposed approach is complementary to the functional-domain-based method for different membrane protein types.     

Prediction of antigenic epitopes on protein surfaces by consensus scoring

Background  

Prediction of antigenic epitopes on protein surfaces is important for vaccine design. Most existing epitope prediction methods focus on protein sequences to predict continuous epitopes linear in sequence. Only a few structure-based epitope prediction algorithms are available and they have not yet shown satisfying performance.     

Ligand supported homology modeling and docking evaluation of CCR2: docked pose selection by consensus scoring

Chemokine receptor 2 (CCR2) is a G-protein coupled receptor (GPCR) and a crucial target for various inflammatory and autoimmune diseases. The structure based antagonists design for many GPCRs, including CCR2, is restricted by the lack of an experimental three dimensional structure. Homology modeling is widely used for the study of GPCR-ligand binding. Since there is substantial diversity for the ligand binding pocket and binding modes among GPCRs, the receptor-ligand binding mode predictions should be derived from homology modeling with supported ligand information. Thus, we modeled the binding of our proprietary CCR2 antagonist using ligand supported homology modeling followed by consensus scoring the docking evaluation based on all modeled binding sites. The protein-ligand model was then validated by visual inspection of receptor-ligand interaction for consistency of published site-directed mutagenesis data and virtual screening a decoy compound database. This model was able to successfully identify active compounds within the decoy database. Finally, additional hit compounds were identified through a docking-based virtual screening of a commercial database, followed by a biological assay to validate CCR2 inhibitory activity. Thus, this procedure can be employed to screen a large database of compounds to identify new CCR2 antagonists.     

Prediction of nucleosome occupancy in Saccharomyces cerevisiae using position-correlation scoring function

Knowledge of the detailed organization of nucleosomes across genomes and the mechanisms of nucleosome positioning is critical for the understanding of gene regulation and expression. In the present work, the bias of 4-mer frequency in nucleosome and linker sequences of the S. cerevisiae genome was analyzed statistically. A novel position-correlation scoring function algorithm based on the bias of 4-mer frequency in linker sequences was presented to distinguish nucleosome vs linker sequences. Five-fold cross-validation demonstrated that the algorithm achieved a good performance with mean area under the receiver operator characteristics curve of 0.981. Next, the algorithm was used to predict nucleosome occupancy throughout the S. cerevisiae genome and relatively high correlation coefficients with experiment maps of nucleosome positioning were obtained. Besides, the distinct nucleosome depleted regions in the vicinity of regulatory sites were confirmed. The results suggest that intrinsic DNA sequence preferences in linker regions have a significant impact on the nucleosome occupancy.     

Computational biotechnology: Prediction of competitive substrate inhibition of enzymes by buffer compounds with protein-ligand docking

In vitro enzymatic activity highly depends on the reaction medium. One of the most important parameters is the buffer used to keep the pH stable. The buffering compound prevents a severe pH-change and therefore a possible denaturation of the enzyme. However buffer agents can also have negative effects on the enzymatic activity, such as competitive substrate inhibition. We assess this effect with a computational approach based on a protein-ligand docking method and the HYDE scoring function. Our method predicts competitive binding of the buffer compound to the active site of the enzyme. Using data from literature and new experimental data, the procedure is evaluated on nine different enzymatic reactions. The method predicts buffer-enzyme interactions and is able to score these interactions with the correct trend of enzymatic activities. Using the new method, possible buffers can be selected or discarded prior to laboratory experiments.     

Prediction of phosphotyrosine signaling networks using a scoring matrix-assisted ligand identification approach

Systematic identification of binding partners for modular domains such as Src homology 2 (SH2) is important for understanding the biological function of the corresponding SH2 proteins. We have developed a worldwide web-accessible computer program dubbed SMALI for scoring matrix-assisted ligand identification for SH2 domains and other signaling modules. The current version of SMALI harbors 76 unique scoring matrices for SH2 domains derived from screening oriented peptide array libraries. These scoring matrices are used to search a protein database for short peptides preferred by an SH2 domain. An experimentally determined cut-off value is used to normalize an SMALI score, therefore allowing for direct comparison in peptide-binding potential for different SH2 domains. SMALI employs distinct scoring matrices from Scansite, a popular motif-scanning program. Moreover, SMALI contains built-in filters for phosphoproteins, Gene Ontology (GO) correlation and colocalization of subject and query proteins. Compared to Scansite, SMALI exhibited improved accuracy in identifying binding peptides for SH2 domains. Applying SMALI to a group of SH2 domains identified hundreds of interactions that overlap significantly with known networks mediated by the corresponding SH2 proteins, suggesting SMALI is a useful tool for facile identification of signaling networks mediated by modular domains that recognize short linear peptide motifs.