Protein contact prediction using patterns of correlation

We describe a new method for using neural networks to predict residue contact pairs in a protein. The main inputs to the neural network are a set of 25 measures of correlated mutation between all pairs of residues in two "windows" of size 5 centered on the residues of interest. While the individual pair-wise correlations are a relatively weak predictor of contact, by training the network on windows of correlation the accuracy of prediction is significantly improved. The neural network is trained on a set of 100 proteins and then tested on a disjoint set of 1033 proteins of known structure. An average predictive accuracy of 21.7% is obtained taking the best L/2 predictions for each protein, where L is the sequence length. Taking the best L/10 predictions gives an average accuracy of 30.7%. The predictor is also tested on a set of 59 proteins from the CASP5 experiment. The accuracy is found to be relatively consistent across different sequence lengths, but to vary widely according to the secondary structure. Predictive accuracy is also found to improve by using multiple sequence alignments containing many sequences to calculate the correlations.     

Experimental analysis of co‐evolution within protein complexes: The yeast exosome as a model

Extensive bioinformatics analysis suggests that the stability and function of protein complexes are maintained throughout evolution by coordinated changes (co‐evolution) of complex subunits. Yet, relatively little is known regarding the actual dynamics of such processes and the functional implications of co‐evolution within protein complexes, since most of the bioinformatics predictions were not analyzed experimentally. Here, we describe a systematic experimental approach that allows a step‐by‐step observation of the co‐evolution process in protein complexes. The exosome complex, an essential complex exhibiting a 3′→5′ RNA degradation activity, served as a model system. In this study, we show that exosome subunits diverged very early during fungal evolution. Interestingly, we found that despite significant differences in conservation between Rrp41 and Mtr3 both subunits exhibit similar divergence pattern and co‐evolutionary behavior through fungi evolution. Activity analysis of mutated exosomes exposes another layer of co‐evolution between the core subunits and RNA substrates. Overall, our approach allows the experimental analysis of co‐evolution within protein complexes and together with bioinformatics analysis can significantly deepen our understanding of the evolution of these complexes. Proteins 2013; 81:1997–2006. © 2013 Wiley Periodicals, Inc.     

Improving accuracy of protein contact prediction using balanced network deconvolution

Residue contact map is essential for protein three‐dimensional structure determination. But most of the current contact prediction methods based on residue co‐evolution suffer from high false‐positives as introduced by indirect and transitive contacts (i.e., residues A–B and B–C are in contact, but A–C are not). Built on the work by Feizi et al. (Nat Biotechnol 2013; 31:726–733), which demonstrated a general network model to distinguish direct dependencies by network deconvolution, this study presents a new balanced network deconvolution (BND) algorithm to identify optimized dependency matrix without limit on the eigenvalue range in the applied network systems. The algorithm was used to filter contact predictions of five widely used co‐evolution methods. On the test of proteins from three benchmark datasets of the 9th critical assessment of protein structure prediction (CASP9), CASP10, and PSICOV (precise structural contact prediction using sparse inverse covariance estimation) database experiments, the BND can improve the medium‐ and long‐range contact predictions at the L/5 cutoff by 55.59% and 47.68%, respectively, without additional central processing unit cost. The improvement is statistically significant, with a P‐value < 5.93 × 10^?3 in the Student's t‐test. A further comparison with the ab initio structure predictions in CASPs showed that the usefulness of the current co‐evolution‐based contact prediction to the three‐dimensional structure modeling relies on the number of homologous sequences existing in the sequence databases. BND can be used as a general contact refinement method, which is freely available at: http://www.csbio.sjtu.edu.cn/bioinf/BND/ . Proteins 2015; 83:485–496. © 2014 Wiley Periodicals, Inc.     

Systematic characterization of pan‐cancer mutation clusters

Cancer genome sequencing has shown that driver genes can often be distinguished not only by the elevated mutation frequency but also by specific nucleotide positions that accumulate changes at a high rate. However, properties associated with a residue's potential to drive tumorigenesis when mutated have not yet been systematically investigated. Here, using a novel methodological approach, we identify and characterize a compendium of 180 hotspot residues within 160 human proteins which occur with a significant frequency and are likely to have functionally relevant impact. We find that such mutations (i) are more prominent in proteins that can exist in the on and off state, (ii) reflect the identity of a tumor of origin, and (iii) often localize within interfaces which mediate interactions with other proteins or ligands. Following, we further examine structural data for human protein complexes and identify a number of additional protein interfaces that accumulate cancer mutations at a high rate. Jointly, these analyses suggest that disruption and dysregulation of protein interactions can be instrumental in switching functions of cancer proteins and activating downstream changes.     

Exploring additivity effects of double mutations on the binding affinity of protein‐protein complexes

Additivity in binding affinity of protein‐protein complexes refers to the change in free energy of binding (ΔΔG_bind) for double (or multiple) mutations which is approximately equal to the sum of their corresponding single mutation ΔΔG_bind values. In this study, we have explored the additivity effect of double mutants, which shows a linear relationship between the binding affinity of double and sum of single mutants with a correlation of 0.90. However, the comparison of ΔΔG_bind values showed a mean absolute deviation of 0.86 kcal/mol, and 25.6% of the double mutants show a deviation of more than 1 kcal/mol, which are identified as non‐additive. The additivity effects have been analyzed based on the influence of structural features such as accessible surface area, long range order, binding propensity change, surrounding hydrophobicity, flexibility, atomic contacts between the mutations and distance between the 2 mutations. We found that non‐additive mutations tend to be closer to each other and have more contacts. We have also used machine learning methods to discriminate additive and non‐additive mutations using structure‐based features, which showed the accuracies in the range of 0.77–0.92 for protein‐protein complexes belonging to different functions. Further, we have compared the additivity effects of protein stability along with binding affinity and explored the similarities and differences between them. The results obtained in this study provide insights into the effects of various structural features on binding affinity of double mutants, and will aid the development of accurate methods to predict the binding affinity of double mutants.     

Correlated mutations: advances and limitations. A study on fusion proteins and on the Cohesin-Dockerin families

Correlated mutations have been repeatedly exploited for intramolecular contact map prediction. Over the last decade these efforts yielded several methods for measuring correlated mutations. Nevertheless, the application of correlated mutations for the prediction of intermolecular interactions has not yet been explored. This gap is due to several obstacles, such as 3D complexes availability, paralog discrimination, and the availability of sequence pairs that are required for inter- but not intramolecular analyses. Here we selected for analysis fusion protein families that bypass some of these obstacles. We find that several correlated mutation measurements yield reasonable accuracy for intramolecular contact map prediction on the fusion dataset. However, the accuracy level drops sharply in intermolecular contacts prediction. This drop in accuracy does not occur always. In the Cohesin-Dockerin family, reasonable accuracy is achieved in the prediction of both intra- and intermolecular contacts. The Cohesin-Dockerin family is well suited for correlated mutation analysis. Because, however, this family constitutes a special case (it has radical mutations, has domain repeats, within each species each Dockerin domain interacts with each Cohesin domain, see below), the successful prediction in this family does not point to a general potential in using correlated mutations for predicting intermolecular contacts. Overall, the results of our study indicate that current methodologies of correlated mutations analysis are not suitable for large-scale intermolecular contact prediction, and thus cannot assist in docking. With current measurements, sequence availability, sequence annotations, and underdeveloped sequence pairing methods, correlated mutations can yield reasonable accuracy only for a handful of families.     

LuTHy: a double‐readout bioluminescence‐based two‐hybrid technology for quantitative mapping of protein–protein interactions in mammalian cells

Information on protein–protein interactions (PPIs) is of critical importance for studying complex biological systems and developing therapeutic strategies. Here, we present a double‐readout bioluminescence‐based two‐hybrid technology, termed LuTHy, which provides two quantitative scores in one experimental procedure when testing binary interactions. PPIs are first monitored in cells by quantification of bioluminescence resonance energy transfer (BRET) and, following cell lysis, are again quantitatively assessed by luminescence‐based co‐precipitation (LuC). The double‐readout procedure detects interactions with higher sensitivity than traditional single‐readout methods and is broadly applicable, for example, for detecting the effects of small molecules or disease‐causing mutations on PPIs. Applying LuTHy in a focused screen, we identified 42 interactions for the presynaptic chaperone CSPα, causative to adult‐onset neuronal ceroid lipofuscinosis (ANCL), a progressive neurodegenerative disease. Nearly 50% of PPIs were found to be affected when studying the effect of the disease‐causing missense mutations L115R and ?L116 in CSPα with LuTHy. Our study presents a robust, sensitive research tool with high utility for investigating the molecular mechanisms by which disease‐associated mutations impair protein activity in biological systems.     

Intrinsic secondary structure propensities of the amino acids,using statistical ϕ–ψ matrices: Comparison with experimental scales

Today there are several different experimental scales for the intrinsic α-helix as well as β-strand, propensities of the 20 amino acids obtained from the thermodynamic analysis of various model systems. These scales do not compare well with those extracted from statistical analysis of three-dimensional structure databases. Possible explanations for this could be the limited size of the databases used, the definitions of intrinsic propensities, or the theoretical approach. Here we report a statistical determination of α-helix and β-strand propensities derived from the analysis of a database of 279 three-dimensional structures. Contrary to what has been generally done, we have considered a particular residue as in α-helix or β-strand conformation by looking only at its dihedral angles (?–ψ matrices). Neither the identity nor the conformation of the surrounding residues in the amino acid sequence has been taken into consideration. Pseudoenergy empirical scales have been calculated from the statistical propensities. These scales agree very well with the experimental ones in relative and absolute terms. Moreover, its correlation with the average of the experimental scales for α-helix or β-strand is as good as the correlations of the individual experimental scales with the average. These results show that by using a large enough database and a proper definition for the secondary structure propensities, it is possible to obtain a scale as good as any of experimental origin. Interestingly the ?–ψ analysis of the Ramachandran plot suggests that the amino acids could have different β-strand propensities in different subregions of the β-strand area. © 1994 Wiley-Liss, Inc.     

Computational design of second‐site suppressor mutations at protein–protein interfaces

The importance of a protein–protein interaction to a signaling pathway can be established by showing that amino acid mutations that weaken the interaction disrupt signaling, and that additional mutations that rescue the interaction recover signaling. Identifying rescue mutations, often referred to as second‐site suppressor mutations, controls against scenarios in which the initial deleterious mutation inactivates the protein or disrupts alternative protein–protein interactions. Here, we test a structure‐based protocol for identifying second‐site suppressor mutations that is based on a strategy previously described by Kortemme and Baker. The molecular modeling software Rosetta is used to scan an interface for point mutations that are predicted to weaken binding but can be rescued by mutations on the partner protein. The protocol typically identifies three types of specificity switches: knob‐in‐to‐hole redesigns, switching hydrophobic interactions to hydrogen bond interactions, and replacing polar interactions with nonpolar interactions. Computational predictions were tested with two separate protein complexes; the G‐protein Gα_i1 bound to the RGS14 GoLoco motif, and UbcH7 bound to the ubiquitin ligase E6AP. Eight designs were experimentally tested. Swapping a buried hydrophobic residue with a polar residue dramatically weakened binding affinities. In none of these cases were we able to identify compensating mutations that returned binding to wild‐type affinity, highlighting the challenges inherent in designing buried hydrogen bond networks. The strongest specificity switches were a knob‐in‐to‐hole design (20‐fold) and the replacement of a charge–charge interaction with nonpolar interactions (55‐fold). In two cases, specificity was further tuned by including mutations distant from the initial design. Proteins 2010. © 2009 Wiley‐Liss, Inc.     

Pathway mapping and development of disease‐specific biomarkers: protein‐based network biomarkers

It is known that a disease is rarely a consequence of an abnormality of a single gene, but reflects the interactions of various processes in a complex network. Annotated molecular networks offer new opportunities to understand diseases within a systems biology framework and provide an excellent substrate for network‐based identification of biomarkers. The network biomarkers and dynamic network biomarkers (DNBs) represent new types of biomarkers with protein–protein or gene–gene interactions that can be monitored and evaluated at different stages and time‐points during development of disease. Clinical bioinformatics as a new way to combine clinical measurements and signs with human tissue‐generated bioinformatics is crucial to translate biomarkers into clinical application, validate the disease specificity, and understand the role of biomarkers in clinical settings. In this article, the recent advances and developments on network biomarkers and DNBs are comprehensively reviewed. How network biomarkers help a better understanding of molecular mechanism of diseases, the advantages and constraints of network biomarkers for clinical application, clinical bioinformatics as a bridge to the development of diseases‐specific, stage‐specific, severity‐specific and therapy predictive biomarkers, and the potentials of network biomarkers are also discussed.     

DockRank: Ranking docked conformations using partner‐specific sequence homology‐based protein interface prediction

Selecting near‐native conformations from the immense number of conformations generated by docking programs remains a major challenge in molecular docking. We introduce DockRank, a novel approach to scoring docked conformations based on the degree to which the interface residues of the docked conformation match a set of predicted interface residues. DockRank uses interface residues predicted by partner‐specific sequence homology‐based protein–protein interface predictor (PS‐HomPPI), which predicts the interface residues of a query protein with a specific interaction partner. We compared the performance of DockRank with several state‐of‐the‐art docking scoring functions using Success Rate (the percentage of cases that have at least one near‐native conformation among the top m conformations) and Hit Rate (the percentage of near‐native conformations that are included among the top m conformations). In cases where it is possible to obtain partner‐specific (PS) interface predictions from PS‐HomPPI, DockRank consistently outperforms both (i) ZRank and IRAD, two state‐of‐the‐art energy‐based scoring functions (improving Success Rate by up to 4‐fold); and (ii) Variants of DockRank that use predicted interface residues obtained from several protein interface predictors that do not take into account the binding partner in making interface predictions (improving success rate by up to 39‐fold). The latter result underscores the importance of using partner‐specific interface residues in scoring docked conformations. We show that DockRank, when used to re‐rank the conformations returned by ClusPro, improves upon the original ClusPro rankings in terms of both Success Rate and Hit Rate. DockRank is available as a server at http://einstein.cs.iastate.edu/DockRank/ . Proteins 2014; 82:250–267. © 2013 Wiley Periodicals, Inc.     

Shaping dots and lines: adding modularity into protein interaction networks using structural information

Determining protein interaction networks and generating models to simulate network changes in time and space are crucial for understanding a biological system and for predicting the effect of mutants found in diseases. In this review we discuss the great potential of using structural information together with computational tools towards reaching this goal: the prediction of new protein interactions, the estimation of affinities and kinetic rate constants between protein complexes, and finally the determination of which interactions are compatible with each other and which interactions are exclusive. The latter one will be important to reorganize large scale networks into functional modular networks.     

Quantifying intracellular equilibrium dissociation constants using single‐channel time‐resolved FRET

Quantification of the intracellular equilibrium dissociation constant of the interaction, K_d, is challenging due to the variability of the relative concentrations of the interacting proteins in the cell. Fluorescence lifetime imaging microscopy (FLIM) of the donor provides an accurate measurement of the molecular fraction of donor involved in FRET, but the fraction of bound acceptor is also needed to reliably estimate K_d. We present a method that exploits the spectroscopic properties of the widely used eGFP – mCherry FRET pair to rigorously determine the intracellular K_d based on imaging the fluorescence lifetime of only the donor (single‐channel FLIM). We have assessed the effect of incomplete labelling and determined its range of application for different K_d using Monte Carlo simulations. We have demonstrated this method estimating the intracellular K_d for the homodimerisaton of the oncogenic protein 3‐phosphoinositide‐dependent kinase 1 (PDK1) in different cell lines and conditions, revealing a competitive mechanism for its regulation. The measured intracellular K_d was validated against in‐vitro data. This method provides an accurate and generic tool to quantify protein interactions in situ.

     

Two‐domain analysis of JmjN‐JmjC and PHD‐JmjC lysine demethylases: Detecting an inter‐domain evolutionary stress

Residues at different positions of a multiple sequence alignment sometimes evolve together, due to a correlated structural or functional stress at these positions. Co‐evolution has thus been evidenced computationally in multiple proteins or protein domains. Here, we wish to study whether an evolutionary stress is exerted on a sequence alignment across protein domains, i.e., on longer sequence separations than within a single protein domain. JmjC‐containing lysine demethylases were chosen for analysis, as a follow‐up to previous studies; these proteins are important multidomain epigenetic regulators. In these proteins, the JmjC domain is responsible for the demethylase activity, and surrounding domains interact with histones, DNA or partner proteins. This family of enzymes was analyzed at the sequence level, in order to determine whether the sequence of JmjC‐domains was affected by the presence of a neighboring JmjN domain or PHD finger in the protein. Multiple positions within JmjC sequences were shown to have their residue distributions significantly altered by the presence of the second domain. Structural considerations confirmed the relevance of the analysis for JmjN‐JmjC proteins, while among PHD‐JmjC proteins, the length of the linker region could be correlated to the residues observed at the most affected positions. The correlation of domain architecture with residue types at certain positions, as well as that of overall architecture with protein function, is discussed. The present results thus evidence the existence of an across‐domain evolutionary stress in JmjC‐containing demethylases, and provide further insights into the overall domain architecture of JmjC domain‐containing proteins.     

Solution properties of γ‐crystallins: Hydration of fish and mammal γ‐crystallins

Lens γ crystallins are found at the highest protein concentration of any tissue, ranging from 300 mg/mL in some mammals to over 1000 mg/mL in fish. Such high concentrations are necessary for the refraction of light, but impose extreme requirements for protein stability and solubility. γ‐crystallins, small stable monomeric proteins, are particularly associated with the lowest hydration regions of the lens. Here, we examine the solvation of selected γ‐crystallins from mammals (human γD and mouse γS) and fish (zebrafish γM2b and γM7). The thermodynamic water binding coefficient B₁ could be probed by sucrose expulsion, and the hydrodynamic hydration shell of tightly bound water was probed by translational diffusion and structure‐based hydrodynamic boundary element modeling. While the amount of tightly bound water of human γD was consistent with that of average proteins, the water binding of mouse γS was found to be relatively low. γM2b and γM7 crystallins were found to exhibit extremely low degrees hydration, consistent with their role in the fish lens. γM crystallins have a very high methionine content, in some species up to 15%. Structure‐based modeling of hydration in γM7 crystallin suggests low hydration is associated with the large number of surface methionine residues, likely in adaptation to the extremely high concentration and low hydration environment in fish lenses. Overall, the degree of hydration appears to balance stability and tissue density requirements required to produce and maintain the optical properties of the lens in different vertebrate species.     

Quantitative analysis and prediction of curvature in leucine‐rich repeat proteins

Leucine‐rich repeat (LRR) proteins form a large and diverse family. They have a wide range of functions most of which involve the formation of protein–protein interactions. All known LRR structures form curved solenoids, although there is large variation in their curvature. It is this curvature that determines the shape and dimensions of the inner space available for ligand binding. Unfortunately, large‐scale parameters such as the overall curvature of a protein domain are extremely difficult to predict. Here, we present a quantitative analysis of determinants of curvature of this family. Individual repeats typically range in length between 20 and 30 residues and have a variety of secondary structures on their convex side. The observed curvature of the LRR domains correlates poorly with the lengths of their individual repeats. We have, therefore, developed a scoring function based on the secondary structure of the convex side of the protein that allows prediction of the overall curvature with a high degree of accuracy. We also demonstrate the effectiveness of this method in selecting a suitable template for comparative modeling. We have developed an automated, quantitative protocol that can be used to predict accurately the curvature of leucine‐rich repeat proteins of unknown structure from sequence alone. This protocol is available as an online resource at http://www.bioinf.manchester.ac.uk/curlrr/ . Proteins 2009. © 2009 Wiley‐Liss, Inc.     

A generalized knowledge‐based discriminatory function for biomolecular interactions

Several novel and established knowledge‐based discriminatory function formulations and reference state derivations have been evaluated to identify parameter sets capable of distinguishing native and near‐native biomolecular interactions from incorrect ones. We developed the r·m·r function, a novel atomic level radial distribution function with mean reference state that averages over all pairwise atom types from a reduced atom type composition, using experimentally determined intermolecular complexes in the Cambridge Structural Database (CSD) and the Protein Data Bank (PDB) as the information sources. We demonstrate that r·m·r had the best discriminatory accuracy and power for protein‐small molecule and protein‐DNA interactions, regardless of whether the native complex was included or excluded, from the test set. The superior performance of the r·m·r discriminatory function compared with seventeen alternative functions evaluated on publicly available test sets for protein‐small molecule and protein‐DNA interactions indicated that the function was not over optimized through back testing on a single class of biomolecular interactions. The initial success of the reduced composition and superior performance with the CSD as the distribution set over the PDB implies that further improvements and generality of the function are possible by deriving probabilities from subsets of the CSD, using structures that consist of only the atom types to be considered for given biomolecular interactions. The method is available as a web server module at http://protinfo.compbio.washington.edu . Proteins 2009. © 2008 Wiley‐Liss, Inc.