Recognizing complex, asymmetric functional sites in protein structures using a Bayesian scoring function

A random forest algorithm-based approach to capture latent decision variables and their cutoff values

Although reference intervals (RIs) and clinical decision limits (CDLs) are vital laboratory information for supporting the interpretation of numerical clinical pathology results, there is evidence that RIs and CDLs vary in certain contexts as well as other evidence that RIs and CDLs are flawed. We propose a random forest algorithm-based exploration methodology by using phenotype transformation of independent variables in relation to dependent variables to capture latent decision variables and their cutoff values. We denote certain CDLs within the RIs estimated by an indirect method that affect some diagnostics or outcomes in the context of specific patients' conditions as latent CDLs. We then apply the proposed methodology to clinical laboratory data regarding bodily fluids, such as blood, urine at the admission of patients for the exploration of latent CDLs of hospital length of stay (HLOS) for each patients' condition identified by diseases of patients who undergoing surgeries. From the exploration results, we found that free Thyroxine (T4) above five unique cutoff values: 1.16 ng/dL, 1.19 ng/dL, 1.2 ng/dL, 1.23 ng/dL and 1.25 ng/dL for tachyarrhythmia predicted longer HLOS, though these cutoff values fall within the estimated RIs as well as the hospital-determined RIs. In addition to the evidence that higher free Thyroxine (T4) levels within the RIs have an association with the corresponding disease, on the whole, the cutoff values except 1.16 ng/dL tended to affect long HLOS with the significant differences. The cutoff values could be taken up for discussion among clinical experts whether it is meaningful to alert the risk of patients' conditions and the long HLOS at the admission of patients. If clinical experts appreciate its meaningfulness in clinical practice, the alerts could be embedded in electronic medical records for handling those risks at the admission of patients.

High precision prediction of functional sites in protein structures

We address the problem of assigning biological function to solved protein structures. Computational tools play a critical role in identifying potential active sites and informing screening decisions for further lab analysis. A critical parameter in the practical application of computational methods is the precision, or positive predictive value. Precision measures the level of confidence the user should have in a particular computed functional assignment. Low precision annotations lead to futile laboratory investigations and waste scarce research resources. In this paper we describe an advanced version of the protein function annotation system FEATURE, which achieved 99% precision and average recall of 95% across 20 representative functional sites. The system uses a Support Vector Machine classifier operating on the microenvironment of physicochemical features around an amino acid. We also compared performance of our method with state-of-the-art sequence-level annotator Pfam in terms of precision, recall and localization. To our knowledge, no other functional site annotator has been rigorously evaluated against these key criteria. The software and predictive models are incorporated into the WebFEATURE service at http://feature.stanford.edu/wf4.0-beta.

TRIAL: a tool for finding distant structural similarities

Finding structural similarities in distantly related proteins can reveal functional relationships that can not be identified using sequence comparison. Given two proteins A and B and threshold ε Å, we develop an algorithm, TRiplet-based Iterative ALignment (TRIAL) for computing the transformation of B that maximizes the number of aligned residues such that the root mean square deviation (RMSD) of the alignment is at most ε Å. Our algorithm is designed with the specific goal of effectively handling proteins with low similarity in primary structure, where existing algorithms perform particularly poorly. Experiments show that our method outperforms existing methods. TRIAL alignment brings the secondary structures of distantly related proteins to similar orientations. It also finds larger number of secondary structure matches at lower RMSD values and increased overall alignment lengths. Its classification accuracy is up to 63 percent better than other methods, including CE and DALI. TRIAL successfully aligns 83 percent of the residues from the smaller protein in reasonable time while other methods align only 29 to 65 percent of the residues for the same set of proteins.

INTEGRATING COMPUTATIONAL PROTEIN FUNCTION PREDICTION INTO DRUG DISCOVERY INITIATIVES

Pharmaceutical researchers must evaluate vast numbers of protein sequences and formulate innovative strategies for identifying valid targets and discovering leads against them as a way of accelerating drug discovery. The ever increasing number and diversity of novel protein sequences identified by genomic sequencing projects and the success of worldwide structural genomics initiatives have spurred great interest and impetus in the development of methods for accurate, computationally empowered protein function prediction and active site identification. Previously, in the absence of direct experimental evidence, homology-based protein function annotation remained the gold-standard for in silico analysis and prediction of protein function. However, with the continued exponential expansion of sequence databases, this approach is not always applicable, as fewer query protein sequences demonstrate significant homology to protein gene products of known function. As a result, several non-homology based methods for protein function prediction that are based on sequence features, structure, evolution, biochemical and genetic knowledge have emerged. Herein, we review current bioinformatic programs and approaches for protein function prediction/annotation and discuss their integration into drug discovery initiatives. The development of such methods to annotate protein functional sites and their application to large protein functional families is crucial to successfully utilizing the vast amounts of genomic sequence information available to drug discovery and development processes.

Graphlet kernels for prediction of functional residues in protein structures

We introduce a novel graph-based kernel method for annotating functional residues in protein structures. A structure is first modeled as a protein contact graph, where nodes correspond to residues and edges connect spatially neighboring residues. Each vertex in the graph is then represented as a vector of counts of labeled non-isomorphic subgraphs (graphlets), centered on the vertex of interest. A similarity measure between two vertices is expressed as the inner product of their respective count vectors and is used in a supervised learning framework to classify protein residues. We evaluated our method on two function prediction problems: identification of catalytic residues in proteins, which is a well-studied problem suitable for benchmarking, and a much less explored problem of predicting phosphorylation sites in protein structures. The performance of the graphlet kernel approach was then compared against two alternative methods, a sequence-based predictor and our implementation of the FEATURE framework. On both tasks, the graphlet kernel performed favorably; however, the margin of difference was considerably higher on the problem of phosphorylation site prediction. While there is data that phosphorylation sites are preferentially positioned in intrinsically disordered regions, we provide evidence that for the sites that are located in structured regions, neither the surface accessibility alone nor the averaged measures calculated from the residue microenvironments utilized by FEATURE were sufficient to achieve high accuracy. The key benefit of the graphlet representation is its ability to capture neighborhood similarities in protein structures via enumerating the patterns of local connectivity in the corresponding labeled graphs.

Prediction of calcium-binding sites by combining loop-modeling with machine learning

Background

Protein ligand-binding sites in the apo state exhibit structural flexibility. This flexibility often frustrates methods for structure-based recognition of these sites because it leads to the absence of electron density for these critical regions, particularly when they are in surface loops. Methods for recognizing functional sites in these missing loops would be useful for recovering additional functional information.

Results

We report a hybrid approach for recognizing calcium-binding sites in disordered regions. Our approach combines loop modeling with a machine learning method (FEATURE) for structure-based site recognition. For validation, we compared the performance of our method on known calcium-binding sites for which there are both holo and apo structures. When loops in the apo structures are rebuilt using modeling methods, FEATURE identifies 14 out of 20 crystallographically proven calcium-binding sites. It only recognizes 7 out of 20 calcium-binding sites in the initial apo crystal structures.

We applied our method to unstructured loops in proteins from SCOP families known to bind calcium in order to discover potential cryptic calcium binding sites. We built 2745 missing loops and evaluated them for potential calcium binding. We made 102 predictions of calcium-binding sites. Ten predictions are consistent with independent experimental verifications. We found indirect experimental evidence for 14 other predictions. The remaining 78 predictions are novel predictions, some with intriguing potential biological significance. In particular, we see an enrichment of beta-sheet folds with predicted calcium binding sites in the connecting loops on the surface that may be important for calcium-mediated function switches.

Conclusion

Protein crystal structures are a potentially rich source of functional information. When loops are missing in these structures, we may be losing important information about binding sites and active sites. We have shown that limited loop modeling (e.g. loops less than 17 residues) combined with pattern matching algorithms can recover functions and propose putative conformations associated with these functions.

Improving structure-based function prediction using molecular dynamics

The number of molecules with solved three-dimensional structure but unknown function is increasing rapidly. Particularly problematic are novel folds with little detectable similarity to molecules of known function. Experimental assays can determine the functions of such molecules, but are time-consuming and expensive. Computational approaches can identify potential functional sites; however, these approaches generally rely on single static structures and do not use information about dynamics. In fact, structural dynamics can enhance function prediction: we coupled molecular dynamics simulations with structure-based function prediction algorithms that identify Ca(2+) binding sites. When applied to 11 challenging proteins, both methods showed substantial improvement in performance, revealing 22 more sites in one case and 12 more in the other, with a modest increase in apparent false positives. Thus, we show that treating molecules as dynamic entities improves the performance of structure-based function prediction methods.

Predicting protein ligand binding sites by combining evolutionary sequence conservation and 3D structure

Identifying a protein's functional sites is an important step towards characterizing its molecular function. Numerous structure- and sequence-based methods have been developed for this problem. Here we introduce ConCavity, a small molecule binding site prediction algorithm that integrates evolutionary sequence conservation estimates with structure-based methods for identifying protein surface cavities. In large-scale testing on a diverse set of single- and multi-chain protein structures, we show that ConCavity substantially outperforms existing methods for identifying both 3D ligand binding pockets and individual ligand binding residues. As part of our testing, we perform one of the first direct comparisons of conservation-based and structure-based methods. We find that the two approaches provide largely complementary information, which can be combined to improve upon either approach alone. We also demonstrate that ConCavity has state-of-the-art performance in predicting catalytic sites and drug binding pockets. Overall, the algorithms and analysis presented here significantly improve our ability to identify ligand binding sites and further advance our understanding of the relationship between evolutionary sequence conservation and structural and functional attributes of proteins. Data, source code, and prediction visualizations are available on the ConCavity web site (http://compbio.cs.princeton.edu/concavity/).

Protein molecules are ubiquitous in the cell; they perform thousands of functions crucial for life. Proteins accomplish nearly all of these functions by interacting with other molecules. These interactions are mediated by specific amino acid positions in the proteins. Knowledge of these “functional sites” is crucial for understanding the molecular mechanisms by which proteins carry out their functions; however, functional sites have not been identified in the vast majority of proteins. Here, we present ConCavity, a computational method that predicts small molecule binding sites in proteins by combining analysis of evolutionary sequence conservation and protein 3D structure. ConCavity provides significant improvement over previous approaches, especially on large, multi-chain proteins. In contrast to earlier methods which only predict entire binding sites, ConCavity makes specific predictions of positions in space that are likely to overlap ligand atoms and of residues that are likely to contact bound ligands. These predictions can be used to aid computational function prediction, to guide experimental protein analysis, and to focus computationally intensive techniques used in drug discovery.

The FEATURE framework for protein function annotation: modeling new functions, improving performance, and extending to novel applications

Structural genomics efforts contribute new protein structures that often lack significant sequence and fold similarity to known proteins. Traditional sequence and structure-based methods may not be sufficient to annotate the molecular functions of these structures. Techniques that combine structural and functional modeling can be valuable for functional annotation. FEATURE is a flexible framework for modeling and recognition of functional sites in macromolecular structures. Here, we present an overview of the main components of the FEATURE framework, and describe the recent developments in its use. These include automating training sets selection to increase functional coverage, coupling FEATURE to structural diversity generating methods such as molecular dynamics simulations and loop modeling methods to improve performance, and using FEATURE in large-scale modeling and structure determination efforts.

Fast dynamics perturbation analysis for prediction of protein functional sites

Background

We present a fast version of the dynamics perturbation analysis (DPA) algorithm to predict functional sites in protein structures. The original DPA algorithm finds regions in proteins where interactions cause a large change in the protein conformational distribution, as measured using the relative entropy D_x. Such regions are associated with functional sites.

Results

The Fast DPA algorithm, which accelerates DPA calculations, is motivated by an empirical observation that D_x in a normal-modes model is highly correlated with an entropic term that only depends on the eigenvalues of the normal modes. The eigenvalues are accurately estimated using first-order perturbation theory, resulting in a N-fold reduction in the overall computational requirements of the algorithm, where N is the number of residues in the protein. The performance of the original and Fast DPA algorithms was compared using protein structures from a standard small-molecule docking test set. For nominal implementations of each algorithm, top-ranked Fast DPA predictions overlapped the true binding site 94% of the time, compared to 87% of the time for original DPA. In addition, per-protein recall statistics (fraction of binding-site residues that are among predicted residues) were slightly better for Fast DPA. On the other hand, per-protein precision statistics (fraction of predicted residues that are among binding-site residues) were slightly better using original DPA. Overall, the performance of Fast DPA in predicting ligand-binding-site residues was comparable to that of the original DPA algorithm.

Conclusion

Compared to the original DPA algorithm, the decreased run time with comparable performance makes Fast DPA well-suited for implementation on a web server and for high-throughput analysis.

The SeqFEATURE library of 3D functional site models: comparison to existing methods and applications to protein function annotation

Structural genomics efforts have led to increasing numbers of novel, uncharacterized protein structures with low sequence identity to known proteins, resulting in a growing need for structure-based function recognition tools. Our method, SeqFEATURE, robustly models protein functions described by sequence motifs using a structural representation. We built a library of models that shows good performance compared to other methods. In particular, SeqFEATURE demonstrates significant improvement over other methods when sequence and structural similarity are low.

Robust recognition of zinc binding sites in proteins

Metals play a variety of roles in biological processes, and hence their presence in a protein structure can yield vital functional information. Because the residues that coordinate a metal often undergo conformational changes upon binding, detection of binding sites based on simple geometric criteria in proteins without bound metal is difficult. However, aspects of the physicochemical environment around a metal binding site are often conserved even when this structural rearrangement occurs. We have developed a Bayesian classifier using known zinc binding sites as positive training examples and nonmetal binding regions that nonetheless contain residues frequently observed in zinc sites as negative training examples. In order to allow variation in the exact positions of atoms, we average a variety of biochemical and biophysical properties in six concentric spherical shells around the site of interest. At a specificity of 99.8%, this method achieves 75.5% sensitivity in unbound proteins at a positive predictive value of 73.6%. We also test its accuracy on predicted protein structures obtained by homology modeling using templates with 30%-50% sequence identity to the target sequences. At a specificity of 99.8%, we correctly identify at least one zinc binding site in 65.5% of modeled proteins. Thus, in many cases, our model is accurate enough to identify metal binding sites in proteins of unknown structure for which no high sequence identity homologs of known structure exist. Both the source code and a Web interface are available to the public at http://feature.stanford.edu/metals.

Clustering protein environments for function prediction: finding PROSITE motifs in 3D

BACKGROUND

Structural genomics initiatives are producing increasing numbers of three-dimensional (3D) structures for which there is little functional information. Structure-based annotation of molecular function is therefore becoming critical. We previously presented FEATURE, a method for describing microenvironments around functional sites in proteins. However, FEATURE uses supervised machine learning and so is limited to building models for sites of known importance and location. We hypothesized that there are a large number of sites in proteins that are associated with function that have not yet been recognized. Toward that end, we have developed a method for clustering protein microenvironments in order to evaluate the potential for discovering novel sites that have not been previously identified.

RESULTS

We have prototyped a computational method for rapid clustering of millions of microenvironments in order to discover residues whose surrounding environments are similar and which may therefore share a functional or structural role. We clustered nearly 2,000,000 environments from 9,600 protein chains and defined 4,550 clusters. As a preliminary validation, we asked whether known 3D environments associated with PROSITE motifs were "rediscovered". We found examples of clusters highly enriched for residues that share PROSITE sequence motifs.

CONCLUSION

Our results demonstrate that we can cluster protein environments successfully using a simplified representation and K-means clustering algorithm. The rediscovery of known 3D motifs allows us to calibrate the size and intercluster distances that characterize useful clusters. This information will then allow us to find new clusters with similar characteristics that represent novel structural or functional sites.

Finding new structural and sequence attributes to predict possible disease association of single amino acid polymorphism (SAP)

MOTIVATION

The rapid accumulation of single amino acid polymorphisms (SAPs), also known as non-synonymous single nucleotide polymorphisms (nsSNPs), brings the opportunities and needs to understand and predict their disease association. Currently published attributes are limited, the detailed mechanisms governing the disease association of a SAP remain unclear and thus, further investigation of new attributes and improvement of the prediction are desired.

RESULTS

A SAP dataset was compiled from the Swiss-Prot variant pages. We extracted and demonstrated the effectiveness of several new biologically informative attributes including the structural neighbor profiles that describe the SAP's microenvironment, nearby functional sites that measure the structure-based and sequence-based distances between the SAP site and its nearby functional sites, aggregation properties that measure the likelihood of protein aggregation and disordered regions that consider whether the SAP is located in structurally disordered regions. The new attributes provided insights into the mechanisms of the disease association of SAPs. We built a support vector machines (SVMs) classifier employing a carefully selected set of new and previously published attributes. Through a strict protein-level 5-fold cross-validation, we attained an overall accuracy of 82.61%, and an MCC of 0.60. Moreover, a web server was developed to provide a user-friendly interface for biologists.

AVAILABILITY

The web server is available at http://sapred.cbi.pku.edu.cn/

Protein function prediction using local 3D templates

The prediction of a protein's function from its 3D structure is becoming more and more important as the worldwide structural genomics initiatives gather pace and continue to solve 3D structures, many of which are of proteins of unknown function. Here, we present a methodology for predicting function from structure that shows great promise. It is based on 3D templates that are defined as specific 3D conformations of small numbers of residues. We use four types of template, covering enzyme active sites, ligand-binding residues, DNA-binding residues and reverse templates. The latter are templates generated from the target structure itself and scanned against a representative subset of all known protein structures. Together, the templates provide a fairly thorough coverage of the known structures and ensure that if there is a match to a known structure it is unlikely to be missed. A new scoring scheme provides a highly sensitive means of discriminating between true positive and false positive template matches. In all, the methodology provides a powerful new tool for function prediction to complement those already in use.

A novel statistical ligand-binding site predictor: application to ATP-binding sites

Structural genomics initiatives are leading to rapid growth in newly determined protein 3D structures, the functional characterization of which may still be inadequate. As an attempt to provide insights into the possible roles of the emerging proteins whose structures are available and/or to complement biochemical research, a variety of computational methods have been developed for the screening and prediction of ligand-binding sites in raw structural data, including statistical pattern classification techniques. In this paper, we report a novel statistical descriptor (the Oriented Shell Model) for protein ligand-binding sites, which utilizes the distance and angular position distribution of various structural and physicochemical features present in immediate proximity to the center of a binding site. Using the support vector machine (SVM) as the classifier, our model identified 69% of the ATP-binding sites in whole-protein scanning tests and in eukaryotic proteins the accuracy is particularly high. We propose that this feature extraction and machine learning procedure can screen out ligand-binding-capable protein candidates and can yield valuable biochemical information for individual proteins.