Support vector machines for predicting rRNA-, RNA-, and DNA-binding proteins from amino acid sequence

Classification of gene function remains one of the most important and demanding tasks in the post-genome era. Most of the current predictive computer methods rely on comparing features that are essentially linear to the protein sequence. However, features of a protein nonlinear to the sequence may also be predictive to its function. Machine learning methods, for instance the Support Vector Machines (SVMs), are particularly suitable for exploiting such features. In this work we introduce SVM and the pseudo-amino acid composition, a collection of nonlinear features extractable from protein sequence, to the field of protein function prediction. We have developed prototype SVMs for binary classification of rRNA-, RNA-, and DNA-binding proteins. Using a protein's amino acid composition and limited range correlation of hydrophobicity and solvent accessible surface area as input, each of the SVMs predicts whether the protein belongs to one of the three classes. In self-consistency and cross-validation tests, which measures the success of learning and prediction, respectively, the rRNA-binding SVM has consistently achieved >95% accuracy. The RNA- and DNA-binding SVMs demonstrate more diverse accuracy, ranging from approximately 76% to approximately 97%. Analysis of the test results suggests the directions of improving the SVMs.     

Prediction of protein relative solvent accessibility with a two-stage SVM approach

Information on relative solvent accessibility (RSA) of amino acid residues in proteins provides valuable clues to the prediction of protein structure and function. A two-stage approach with support vector machines (SVMs) is proposed, where an SVM predictor is introduced to the output of the single-stage SVM approach to take into account the contextual relationships among solvent accessibilities for the prediction. By using the position-specific scoring matrices (PSSMs) generated by PSI-BLAST, the two-stage SVM approach achieves accuracies up to 90.4% and 90.2% on the Manesh data set of 215 protein structures and the RS126 data set of 126 nonhomologous globular proteins, respectively, which are better than the highest published scores on both data sets to date. A Web server for protein RSA prediction using a two-stage SVM method has been developed and is available (http://birc.ntu.edu.sg/~pas0186457/rsa.html).     

Residue-level prediction of DNA-binding sites and its application on DNA-binding protein predictions

Protein-DNA interactions are crucial to many cellular activities such as expression-control and DNA-repair. These interactions between amino acids and nucleotides are highly specific and any aberrance at the binding site can render the interaction completely incompetent. In this study, we have three aims focusing on DNA-binding residues on the protein surface: to develop an automated approach for fast and reliable recognition of DNA-binding sites; to improve the prediction by distance-dependent refinement; use these predictions to identify DNA-binding proteins. We use a support vector machines (SVM)-based approach to harness the features of the DNA-binding residues to distinguish them from non-binding residues. Features used for distinction include the residue's identity, charge, solvent accessibility, average potential, the secondary structure it is embedded in, neighboring residues, and location in a cationic patch. These features collected from 50 proteins are used to train SVM. Testing is then performed on another set of 37 proteins, much larger than any testing set used in previous studies. The testing set has no more than 20% sequence identity not only among its pairs, but also with the proteins in the training set, thus removing any undesired redundancy due to homology. This set also has proteins with an unseen DNA-binding structural class not present in the training set. With the above features, an accuracy of 66% with balanced sensitivity and specificity is achieved without relying on homology or evolutionary information. We then develop a post-processing scheme to improve the prediction using the relative location of the predicted residues. Balanced success is then achieved with average sensitivity, specificity and accuracy pegged at 71.3%, 69.3% and 70.5%, respectively. Average net prediction is also around 70%. Finally, we show that the number of predicted DNA-binding residues can be used to differentiate DNA-binding proteins from non-DNA-binding proteins with an accuracy of 78%. Results presented here demonstrate that machine-learning can be applied to automated identification of DNA-binding residues and that the success rate can be ameliorated as more features are added. Such functional site prediction protocols can be useful in guiding consequent works such as site-directed mutagenesis and macromolecular docking.     

Using electrostatic potentials to predict DNA-binding sites on DNA-binding proteins

A method to detect DNA-binding sites on the surface of a protein structure is important for functional annotation. This work describes the analysis of residue patches on the surface of DNA-binding proteins and the development of a method of predicting DNA-binding sites using a single feature of these surface patches. Surface patches and the DNA-binding sites were initially analysed for accessibility, electrostatic potential, residue propensity, hydrophobicity and residue conservation. From this, it was observed that the DNA-binding sites were, in general, amongst the top 10% of patches with the largest positive electrostatic scores. This knowledge led to the development of a prediction method in which patches of surface residues were selected such that they excluded residues with negative electrostatic scores. This method was used to make predictions for a data set of 56 non-homologous DNA-binding proteins. Correct predictions made for 68% of the data set.     

High-throughput prediction of RNA,DNA and protein binding regions mediated by intrinsic disorder

Intrinsically disordered proteins and regions (IDPs and IDRs) lack stable 3D structure under physiological conditions in-vitro, are common in eukaryotes, and facilitate interactions with RNA, DNA and proteins. Current methods for prediction of IDPs and IDRs do not provide insights into their functions, except for a handful of methods that address predictions of protein-binding regions. We report first-of-its-kind computational method DisoRDPbind for high-throughput prediction of RNA, DNA and protein binding residues located in IDRs from protein sequences. DisoRDPbind is implemented using a runtime-efficient multi-layered design that utilizes information extracted from physiochemical properties of amino acids, sequence complexity, putative secondary structure and disorder and sequence alignment. Empirical tests demonstrate that it provides accurate predictions that are competitive with other predictors of disorder-mediated protein binding regions and complementary to the methods that predict RNA- and DNA-binding residues annotated based on crystal structures. Application in Homo sapiens, Mus musculus, Caenorhabditis elegans and Drosophila melanogaster proteomes reveals that RNA- and DNA-binding proteins predicted by DisoRDPbind complement and overlap with the corresponding known binding proteins collected from several sources. Also, the number of the putative protein-binding regions predicted with DisoRDPbind correlates with the promiscuity of proteins in the corresponding protein–protein interaction networks. Webserver: http://biomine.ece.ualberta.ca/DisoRDPbind/     

Function prediction for DNA-/RNA-binding proteins, GPCRs, and drug ADME-associated proteins by SVM

This paper explores the use of support vector machine (SVM) for protein function prediction. Studies are conducted on several groups of proteins with different functions including DNA-binding proteins, RNA-binding proteins, G-protein coupled receptors, drug absorption proteins, drug metabolizing enzymes, drug distribution and excretion proteins. The computed accuracy for the prediction of these proteins is found to be in the range of 82.32% to 99.7%, which illustrates the potential of SVM in facilitating protein function prediction.     

Reduced amino acid alphabet is sufficient to accurately recognize intrinsically disordered protein

Intrinsically disordered proteins are an important class of proteins with unique functions and properties. Here, we have applied a support vector machine (SVM) trained on naturally occurring disordered and ordered proteins to examine the contribution of various parameters (vectors) to recognizing proteins that contain disordered regions. We find that a SVM that incorporates only amino acid composition has a recognition accuracy of 87+/-2%. This result suggests that composition alone is sufficient to accurately recognize disorder. Interestingly, SVMs using reduced sets of amino acids based on chemical similarity preserve high recognition accuracy. A set as small as four retains an accuracy of 84+/-2%; this suggests that general physicochemical properties rather than specific amino acids are important factors contributing to protein disorder.     

Secondary structure prediction with support vector machines

MOTIVATION: A new method that uses support vector machines (SVMs) to predict protein secondary structure is described and evaluated. The study is designed to develop a reliable prediction method using an alternative technique and to investigate the applicability of SVMs to this type of bioinformatics problem. METHODS: Binary SVMs are trained to discriminate between two structural classes. The binary classifiers are combined in several ways to predict multi-class secondary structure. RESULTS: The average three-state prediction accuracy per protein (Q(3)) is estimated by cross-validation to be 77.07 +/- 0.26% with a segment overlap (Sov) score of 73.32 +/- 0.39%. The SVM performs similarly to the 'state-of-the-art' PSIPRED prediction method on a non-homologous test set of 121 proteins despite being trained on substantially fewer examples. A simple consensus of the SVM, PSIPRED and PROFsec achieves significantly higher prediction accuracy than the individual methods.     

SVMyr: A Web Server Detecting Co- and Post-translational Myristoylation in Proteins

Myristoylation (MYR) is a protein modification where a myristoyl group is covalently attached to an exposed (N-terminal) glycine residue. Glycine myristoylation occurs during protein translation (co-translation) or after (post-translation). Myristoylated proteins have a role in signal transduction, apoptosis, and pathogen-mediated processes and their prediction can help in functionally annotating the fraction of proteins undergoing MYR in different proteomes.Here we present SVMyr, a web server allowing the detection of both co- and post-translational myristoylation sites, based on Support Vector Machines (SVM). The input encodes composition and physicochemical features of the octapeptides, known to act as substrates and to physically interact with N-myristoyltransferases (NMTs), the enzymes catalyzing the myristoylation reaction.The method, adopting a cross validation procedure, scores with values of Area Under the Curve (AUC) and Matthews Correlation Coefficient (MCC) of 0.92 and 0.61, respectively. When benchmarked on an independent dataset including experimentally detected 88 medium/high confidence co-translational myristoylation sites and 528 negative examples, SVMyr outperforms available methods, with AUC and MCC equal to 0.91 and 0.58, respectively.A unique feature of SVMyr is the ability to predict post-translational myristoylation sites by coupling the trained SVMs with the detection of caspase cleavage sites, identified by searching regular motifs matching upstream caspase cleavage sites, as reported in literature.Finally, SVMyr confirms 96% of the UniProt set of the electronically annotated myristoylated proteins (31,048) and identifies putative myristoylomes in eight different proteomes, highlighting also new putative NMT substrates.SVMyr is freely available through a user-friendly web server at https://busca.biocomp.unibo.it/lipipred.     

Prediction of RNA-binding proteins from primary sequence by a support vector machine approach

Elucidation of the interaction of proteins with different molecules is of significance in the understanding of cellular processes. Computational methods have been developed for the prediction of protein-protein interactions. But insufficient attention has been paid to the prediction of protein-RNA interactions, which play central roles in regulating gene expression and certain RNA-mediated enzymatic processes. This work explored the use of a machine learning method, support vector machines (SVM), for the prediction of RNA-binding proteins directly from their primary sequence. Based on the knowledge of known RNA-binding and non-RNA-binding proteins, an SVM system was trained to recognize RNA-binding proteins. A total of 4011 RNA-binding and 9781 non-RNA-binding proteins was used to train and test the SVM classification system, and an independent set of 447 RNA-binding and 4881 non-RNA-binding proteins was used to evaluate the classification accuracy. Testing results using this independent evaluation set show a prediction accuracy of 94.1%, 79.3%, and 94.1% for rRNA-, mRNA-, and tRNA-binding proteins, and 98.7%, 96.5%, and 99.9% for non-rRNA-, non-mRNA-, and non-tRNA-binding proteins, respectively. The SVM classification system was further tested on a small class of snRNA-binding proteins with only 60 available sequences. The prediction accuracy is 40.0% and 99.9% for snRNA-binding and non-snRNA-binding proteins, indicating a need for a sufficient number of proteins to train SVM. The SVM classification systems trained in this work were added to our Web-based protein functional classification software SVMProt, at http://jing.cz3.nus.edu.sg/cgi-bin/svmprot.cgi. Our study suggests the potential of SVM as a useful tool for facilitating the prediction of protein-RNA interactions.     

COMSAT: Residue contact prediction of transmembrane proteins based on support vector machines and mixed integer linear programming

In this article, we present COMSAT, a hybrid framework for residue contact prediction of transmembrane (TM) proteins, integrating a support vector machine (SVM) method and a mixed integer linear programming (MILP) method. COMSAT consists of two modules: COMSAT_SVM which is trained mainly on position–specific scoring matrix features, and COMSAT_MILP which is an ab initio method based on optimization models. Contacts predicted by the SVM model are ranked by SVM confidence scores, and a threshold is trained to improve the reliability of the predicted contacts. For TM proteins with no contacts above the threshold, COMSAT_MILP is used. The proposed hybrid contact prediction scheme was tested on two independent TM protein sets based on the contact definition of 14 Å between Cα‐Cα atoms. First, using a rigorous leave‐one‐protein‐out cross validation on the training set of 90 TM proteins, an accuracy of 66.8%, a coverage of 12.3%, a specificity of 99.3% and a Matthews' correlation coefficient (MCC) of 0.184 were obtained for residue pairs that are at least six amino acids apart. Second, when tested on a test set of 87 TM proteins, the proposed method showed a prediction accuracy of 64.5%, a coverage of 5.3%, a specificity of 99.4% and a MCC of 0.106. COMSAT shows satisfactory results when compared with 12 other state‐of‐the‐art predictors, and is more robust in terms of prediction accuracy as the length and complexity of TM protein increase. COMSAT is freely accessible at http://hpcc.siat.ac.cn/COMSAT/ . Proteins 2016; 84:332–348. © 2016 Wiley Periodicals, Inc.     

Enzyme family classification by support vector machines

One approach for facilitating protein function prediction is to classify proteins into functional families. Recent studies on the classification of G-protein coupled receptors and other proteins suggest that a statistical learning method, Support vector machines (SVM), may be potentially useful for protein classification into functional families. In this work, SVM is applied and tested on the classification of enzymes into functional families defined by the Enzyme Nomenclature Committee of IUBMB. SVM classification system for each family is trained from representative enzymes of that family and seed proteins of Pfam curated protein families. The classification accuracy for enzymes from 46 families and for non-enzymes is in the range of 50.0% to 95.7% and 79.0% to 100% respectively. The corresponding Matthews correlation coefficient is in the range of 54.1% to 96.1%. Moreover, 80.3% of the 8,291 correctly classified enzymes are uniquely classified into a specific enzyme family by using a scoring function, indicating that SVM may have certain level of unique prediction capability. Testing results also suggest that SVM in some cases is capable of classification of distantly related enzymes and homologous enzymes of different functions. Effort is being made to use a more comprehensive set of enzymes as training sets and to incorporate multi-class SVM classification systems to further enhance the unique prediction accuracy. Our results suggest the potential of SVM for enzyme family classification and for facilitating protein function prediction. Our software is accessible at http://jing.cz3.nus.edu.sg/cgi-bin/svmprot.cgi.     

Enhanced membrane protein topology prediction using a hierarchical classification method and a new scoring function

The prediction of transmembrane (TM) helix and topology provides important information about the structure and function of a membrane protein. Due to the experimental difficulties in obtaining a high-resolution model, computational methods are highly desirable. In this paper, we present a hierarchical classification method using support vector machines (SVMs) that integrates selected features by capturing the sequence-to-structure relationship and developing a new scoring function based on membrane protein folding. The proposed approach is evaluated on low- and high-resolution data sets with cross-validation, and the topology (sidedness) prediction accuracy reaches as high as 90%. Our method is also found to correctly predict both the location of TM helices and the topology for 69% of the low-resolution benchmark set. We also test our method for discrimination between soluble and membrane proteins and achieve very low overall false positive (0.5%) and false negative rates (0 to approximately 1.2%). Lastly, the analysis of the scoring function suggests that the topogeneses of single-spanning and multispanning TM proteins have different levels of complexity, and the consideration of interloop topogenic interactions for the latter is the key to achieving better predictions. This method can facilitate the annotation of membrane proteomes to extract useful structural and functional information. It is publicly available at http://bio-cluster.iis.sinica.edu.tw/~bioapp/SVMtop.     

Reliability measures for membrane protein topology prediction algorithms

We have developed reliability scores for five widely used membrane protein topology prediction methods, and have applied them both on a test set of 92 bacterial plasma membrane proteins with experimentally determined topologies and on all predicted helix bundle membrane proteins in three fully sequenced genomes: Escherichia coli, Saccharomyces cerevisiae and Caenorhabditis elegans. We show that the reliability scores work well for the TMHMM and MEMSAT methods, and that they allow the probability that the predicted topology is correct to be estimated for any protein. We further show that the available test set is biased towards high-scoring proteins when compared to the genome-wide data sets, and provide estimates for the expected prediction accuracy of TMHMM across the three genomes. Finally, we show that the performance of TMHMM is considerably better when limited experimental information (such as the in/out location of a protein's C terminus) is available, and estimate that at least ten percentage points in overall accuracy in whole-genome predictions can be gained in this way.     

Hierarchical multi-label prediction of gene function

MOTIVATION: Assigning functions for unknown genes based on diverse large-scale data is a key task in functional genomics. Previous work on gene function prediction has addressed this problem using independent classifiers for each function. However, such an approach ignores the structure of functional class taxonomies, such as the Gene Ontology (GO). Over a hierarchy of functional classes, a group of independent classifiers where each one predicts gene membership to a particular class can produce a hierarchically inconsistent set of predictions, where for a given gene a specific class may be predicted positive while its inclusive parent class is predicted negative. Taking the hierarchical structure into account resolves such inconsistencies and provides an opportunity for leveraging all classifiers in the hierarchy to achieve higher specificity of predictions. RESULTS: We developed a Bayesian framework for combining multiple classifiers based on the functional taxonomy constraints. Using a hierarchy of support vector machine (SVM) classifiers trained on multiple data types, we combined predictions in our Bayesian framework to obtain the most probable consistent set of predictions. Experiments show that over a 105-node subhierarchy of the GO, our Bayesian framework improves predictions for 93 nodes. As an additional benefit, our method also provides implicit calibration of SVM margin outputs to probabilities. Using this method, we make function predictions for multiple proteins, and experimentally confirm predictions for proteins involved in mitosis. SUPPLEMENTARY INFORMATION: Results for the 105 selected GO classes and predictions for 1059 unknown genes are available at: http://function.princeton.edu/genesite/ CONTACT: ogt@cs.princeton.edu.     

A new representation for protein secondary structure prediction based on frequent patterns

MOTIVATION: A new representation for protein secondary structure prediction based on frequent amino acid patterns is described and evaluated. We discuss in detail how to identify frequent patterns in a protein sequence database using a level-wise search technique, how to define a set of features from those patterns and how to use those features in the prediction of the secondary structure of a protein sequence using support vector machines (SVMs). RESULTS: Three different sets of features based on frequent patterns are evaluated in a blind testing setup using 150 targets from the EVA contest and compared to predictions of PSI-PRED, PHD and PROFsec. Despite being trained on only 940 proteins, a simple SVM classifier based on this new representation yields results comparable to PSI-PRED and PROFsec. Finally, we show that the method contributes significant information to consensus predictions. AVAILABILITY: The method is available from the authors upon request.     

On the Relation Between the Predicted Secondary Structure and the Protein Size

Accurately predicted protein secondary structure provides useful information for target selection, to analyze protein function and to predict higher dimensional structure. Existing research shows that more data + refined search = better prediction. We analyze relation between the prediction accuracy and another crucial factor, the protein size. Empirical tests performed with two secondary structure predictors on a large set of high-resolution, non-redundant proteins show that the average accuracies for small proteins (<100 residues) equal 73% and 54% for alpha-helices and beta-strands, respectively. The alpha-helix/beta-strand accuracies for very large proteins (>300 residues) equal 77%/68%, respectively. Similarly, the tests with three secondary structure content predictors show that the prediction errors for the small/very large proteins equal 0.13/0.09 and 0.09/0.06 for alpha-helix and beta-strand content, respectively. Our tests confirm that the secondary structure/content predictions for the very large proteins are characterized statistically significantly better quality than prediction for the small proteins. This is in contrast with the tertiary structure predictions in which higher accuracy is obtained for smaller proteins.     

Kernel-based machine learning protocol for predicting DNA-binding proteins

DNA-binding proteins (DNA-BPs) play a pivotal role in various intra- and extra-cellular activities ranging from DNA replication to gene expression control. Attempts have been made to identify DNA-BPs based on their sequence and structural information with moderate accuracy. Here we develop a machine learning protocol for the prediction of DNA-BPs where the classifier is Support Vector Machines (SVMs). Information used for classification is derived from characteristics that include surface and overall composition, overall charge and positive potential patches on the protein surface. In total 121 DNA-BPs and 238 non-binding proteins are used to build and evaluate the protocol. In self-consistency, accuracy value of 100% has been achieved. For cross-validation (CV) optimization over entire dataset, we report an accuracy of 90%. Using leave 1-pair holdout evaluation, the accuracy of 86.3% has been achieved. When we restrict the dataset to less than 20% sequence identity amongst the proteins, the holdout accuracy is achieved at 85.8%. Furthermore, seven DNA-BPs with unbounded structures are all correctly predicted. The current performances are better than results published previously. The higher accuracy value achieved here originates from two factors: the ability of the SVM to handle features that demonstrate a wide range of discriminatory power and, a different definition of the positive patch. Since our protocol does not lean on sequence or structural homology, it can be used to identify or predict proteins with DNA-binding function(s) regardless of their homology to the known ones.     

Mimicking cellular sorting improves prediction of subcellular localization

Predicting the native subcellular compartment of a protein is an important step toward elucidating its function. Here we introduce LOCtree, a hierarchical system combining support vector machines (SVMs) and other prediction methods. LOCtree predicts the subcellular compartment of a protein by mimicking the mechanism of cellular sorting and exploiting a variety of sequence and predicted structural features in its input. Currently LOCtree does not predict localization for membrane proteins, since the compositional properties of membrane proteins significantly differ from those of non-membrane proteins. While any information about function can be used by the system, we present estimates of performance that are valid when only the amino acid sequence of a protein is known. When evaluated on a non-redundant test set, LOCtree achieved sustained levels of 74% accuracy for non-plant eukaryotes, 70% for plants, and 84% for prokaryotes. We rigorously benchmarked LOCtree in comparison to the best alternative methods for localization prediction. LOCtree outperformed all other methods in nearly all benchmarks. Localization assignments using LOCtree agreed quite well with data from recent large-scale experiments. Our preliminary analysis of a few entirely sequenced organisms, namely human (Homo sapiens), yeast (Saccharomyces cerevisiae), and weed (Arabidopsis thaliana) suggested that over 35% of all non-membrane proteins are nuclear, about 20% are retained in the cytosol, and that every fifth protein in the weed resides in the chloroplast.     

Prediction of the functional class of lipid binding proteins from sequence-derived properties irrespective of sequence similarity

Lipid binding proteins play important roles in signaling, regulation, membrane trafficking, immune response, lipid metabolism, and transport. Because of their functional and sequence diversity, it is desirable to explore additional methods for predicting lipid binding proteins irrespective of sequence similarity. This work explores the use of support vector machines (SVMs) as such a method. SVM prediction systems are developed using 14,776 lipid binding and 133,441 nonlipid binding proteins and are evaluated by an independent set of 6,768 lipid binding and 64,761 nonlipid binding proteins. The computed prediction accuracy is 78.9, 79.5, 82.2, 79.5, 84.4, 76.6, 90.6, 79.0, and 89.9% for lipid degradation, lipid metabolism, lipid synthesis, lipid transport, lipid binding, lipopolysaccharide biosynthesis, lipoprotein, lipoyl, and all lipid binding proteins, respectively. The accuracy for the nonmember proteins of each class is 99.9, 99.2, 99.6, 99.8, 99.9, 99.8, 98.5, 99.9, and 97.0%, respectively. Comparable accuracies are obtained when homologous proteins are considered as one, or by using a different SVM kernel function. Our method predicts 86.8% of the 76 lipid binding proteins nonhomologous to any protein in the Swiss-Prot database and 89.0% of the 73 known lipid binding domains as lipid binding. These findings suggest the usefulness of SVMs for facilitating the prediction of lipid binding proteins. Our software can be accessed at the SVMProt server (http://jing.cz3.nus.edu.sg/cgi-bin/svmprot.cgi).