Classification of protein families and detection of the determinant residues with an improved self-organizing map

 Using a SOM (self-organizing map) we can classify sequences within a protein family into subgroups that generally correspond to biological subcategories. These maps tend to show sequence similarity as proximity in the map. Combining maps generated at different levels of resolution, the structure of relations in protein families can be captured that could not otherwise be represented in a single map. The underlying representation of maps enables us to retrieve characteristic sequence patterns for individual subgroups of sequences. Such patterns tend to correspond to functionally important regions. We present a modified SOM algorithm that includes a convergence test that dynamically controls the learning parameters to adapt them to the learning set instead of being fixed and externally optimized by trial and error. Given the variability of protein family size and distribution, the addition of this feature is necessary. The method is successfully tested with a number of families. The rab family of small GTPases is used to illustrate the performance of the method. Received: 25 July 1996 / Accepted in revised form: 13 February 1997     

A protein map and its application

Graphical representation of gene sequences provides a simple way of viewing, sorting, and comparing various gene structures. Here we first report a two-dimensional graphical representation for protein sequences. With this method, we constructed the moment vectors for protein sequences, and mathematically proved that the correspondence between moment vectors and protein sequences is one-to-one. Therefore, each protein sequence can be represented as a point in a map, which we call protein map, and cluster analysis can be used for comparison between the points. Sixty-six proteins from five protein families were analyzed using this method. Our data showed that for proteins in the same family, their corresponding points in the map are close to each other. We also illustrate the efficiency of this approach by performing an extensive cluster analysis of the protein kinase C family. These results indicate that this protein map could be used to mathematically specify the similarity of two proteins and predict properties of an unknown protein based on its amino acid sequence.     

A hypergraph-based method for unification of existing protein structure- and sequence-families

Classification of proteins is a major challenge in bioinformatics. Here an approach is presented, that unifies different existing classifications of protein structures and sequences. Protein structural domains are represented as nodes in a hypergraph. Shared memberships in sequence families result in hyperedges in the graph. The presented method partitions the hypergraph into clusters of structural domains. Each computed cluster is based on a set of shared sequence family memberships. Thus, the clusters put existing protein sequence families into the context of structural family hierarchies. Conversely, structural domains are related to their sequence family memberships, which can be used to gain further knowledge about the respective structural families.     

Predicting functional family of novel enzymes irrespective of sequence similarity: a statistical learning approach

The function of a protein that has no sequence homolog of known function is difficult to assign on the basis of sequence similarity. The same problem may arise for homologous proteins of different functions if one is newly discovered and the other is the only known protein of similar sequence. It is desirable to explore methods that are not based on sequence similarity. One approach is to assign functional family of a protein to provide useful hint about its function. Several groups have employed a statistical learning method, support vector machines (SVMs), for predicting protein functional family directly from sequence irrespective of sequence similarity. These studies showed that SVM prediction accuracy is at a level useful for functional family assignment. But its capability for assignment of distantly related proteins and homologous proteins of different functions has not been critically and adequately assessed. Here SVM is tested for functional family assignment of two groups of enzymes. One consists of 50 enzymes that have no homolog of known function from PSI-BLAST search of protein databases. The other contains eight pairs of homologous enzymes of different families. SVM correctly assigns 72% of the enzymes in the first group and 62% of the enzyme pairs in the second group, suggesting that it is potentially useful for facilitating functional study of novel proteins. A web version of our software, SVMProt, is accessible at http://jing.cz3.nus.edu.sg/cgi-bin/svmprot.cgi.     

A comparison of sequence and structure protein domain families as a basis for structural genomics.

MOTIVATION: Protein families can be defined based on structure or sequence similarity. We wanted to compare two protein family databases, one based on structural and one on sequence similarity, to investigate to what extent they overlap, the similarity in definition of corresponding families, and to create a list of large protein families with unknown structure as a resource for structural genomics. We also wanted to increase the sensitivity of fold assignment by exploiting protein family HMMs. RESULTS: We compared Pfam, a protein family database based on sequence similarity, to Scop, which is based on structural similarity. We found that 70% of the Scop families exist in Pfam while 57% of the Pfam families exist in Scop. Most families that occur in both databases correspond well to each other, but in some cases they are different. Such cases highlight situations in which structure and sequence approaches differ significantly. The comparison enabled us to compile a list of the largest families that do not occur in Scop; these are suitable targets for structure prediction and determination, and may be useful to guide projects in structural genomics. It can be noted that 13 out of the 20 largest protein families without a known structure are likely transmembrane proteins. We also exploited Pfam to increase the sensitivity of detecting homologs of proteins with known structure, by comparing query sequences to Pfam HMMs that correspond to Scop families. For SWISSPROT+TREMBL, this yielded an increase in fold assignment from 31% to 42% compared to using FASTA only. This method assigned a structure to 22% of the proteins in Saccharomyces cerevisiae, 24% in Escherichia coli, and 16% in Methanococcus jannaschii.     

Continuous Distributed Representation of Biological Sequences for Deep Proteomics and Genomics

We introduce a new representation and feature extraction method for biological sequences. Named bio-vectors (BioVec) to refer to biological sequences in general with protein-vectors (ProtVec) for proteins (amino-acid sequences) and gene-vectors (GeneVec) for gene sequences, this representation can be widely used in applications of deep learning in proteomics and genomics. In the present paper, we focus on protein-vectors that can be utilized in a wide array of bioinformatics investigations such as family classification, protein visualization, structure prediction, disordered protein identification, and protein-protein interaction prediction. In this method, we adopt artificial neural network approaches and represent a protein sequence with a single dense n-dimensional vector. To evaluate this method, we apply it in classification of 324,018 protein sequences obtained from Swiss-Prot belonging to 7,027 protein families, where an average family classification accuracy of 93%±0.06% is obtained, outperforming existing family classification methods. In addition, we use ProtVec representation to predict disordered proteins from structured proteins. Two databases of disordered sequences are used: the DisProt database as well as a database featuring the disordered regions of nucleoporins rich with phenylalanine-glycine repeats (FG-Nups). Using support vector machine classifiers, FG-Nup sequences are distinguished from structured protein sequences found in Protein Data Bank (PDB) with a 99.8% accuracy, and unstructured DisProt sequences are differentiated from structured DisProt sequences with 100.0% accuracy. These results indicate that by only providing sequence data for various proteins into this model, accurate information about protein structure can be determined. Importantly, this model needs to be trained only once and can then be applied to extract a comprehensive set of information regarding proteins of interest. Moreover, this representation can be considered as pre-training for various applications of deep learning in bioinformatics. The related data is available at Life Language Processing Website: http://llp.berkeley.edu and Harvard Dataverse: http://dx.doi.org/10.7910/DVN/JMFHTN.     

An efficient algorithm for large-scale detection of protein families

Detection of protein families in large databases is one of the principal research objectives in structural and functional genomics. Protein family classification can significantly contribute to the delineation of functional diversity of homologous proteins, the prediction of function based on domain architecture or the presence of sequence motifs as well as comparative genomics, providing valuable evolutionary insights. We present a novel approach called TRIBE-MCL for rapid and accurate clustering of protein sequences into families. The method relies on the Markov cluster (MCL) algorithm for the assignment of proteins into families based on precomputed sequence similarity information. This novel approach does not suffer from the problems that normally hinder other protein sequence clustering algorithms, such as the presence of multi-domain proteins, promiscuous domains and fragmented proteins. The method has been rigorously tested and validated on a number of very large databases, including SwissProt, InterPro, SCOP and the draft human genome. Our results indicate that the method is ideally suited to the rapid and accurate detection of protein families on a large scale. The method has been used to detect and categorise protein families within the draft human genome and the resulting families have been used to annotate a large proportion of human proteins.     

Protein-DNA interactions: amino acid conservation and the effects of mutations on binding specificity

We investigate the conservation of amino acid residue sequences in 21 DNA-binding protein families and study the effects that mutations have on DNA-sequence recognition. The observations are best understood by assigning each protein family to one of three classes: (i) non-specific, where binding is independent of DNA sequence; (ii) highly specific, where binding is specific and all members of the family target the same DNA sequence; and (iii) multi-specific, where binding is also specific, but individual family members target different DNA sequences. Overall, protein residues in contact with the DNA are better conserved than the rest of the protein surface, but there is a complex underlying trend of conservation for individual residue positions. Amino acid residues that interact with the DNA backbone are well conserved across all protein families and provide a core of stabilising contacts for homologous protein-DNA complexes. In contrast, amino acid residues that interact with DNA bases have variable levels of conservation depending on the family classification. In non-specific families, base-contacting residues are well conserved and interactions are always found in the minor groove where there is little discrimination between base types. In highly specific families, base-contacting residues are highly conserved and allow member proteins to recognise the same target sequence. In multi-specific families, base-contacting residues undergo frequent mutations and enable different proteins to recognise distinct target sequences. Finally, we report that interactions with bases in the target sequence often follow (though not always) a universal code of amino acid-base recognition and the effects of amino acid mutations can be most easily understood for these interactions.     

A unifold, mesofold, and superfold model of protein fold use.

As more and more protein structures are determined, there is increasing interest in the question of how many different folds have been used in biology. The history of the rate of discovery of new folds and the distribution of sequence families among known folds provide a means of estimating the underlying distribution of fold use. Previous models exploiting these data have led to rather different conclusions on the total number of folds. We present a new model, based on the notion that the folds used in biology fall naturally into three classes: unifolds, that is, folds found only in a single narrow sequence family; mesofolds, found in an intermediate number of families; and the previously noted superfolds, found in many protein families. We show that this model fits the available data well and has predicted the development of SCOP over the past 2 years. The principle implications of the model are as follows: (1) The vast majority of folds will be found in only a single sequence family; (2) the total number of folds is at least 10,000; and (3) 80% of sequence families have one of about 400 folds, most of which are already known.     

Automatic methods for predicting functionally important residues

Sequence analysis is often the first guide for the prediction of residues in a protein family that may have functional significance. A few methods have been proposed which use the division of protein families into subfamilies in the search for those positions that could have some functional significance for the whole family, but at the same time which exhibit the specificity of each subfamily ("Tree-determinant residues"). However, there are still many unsolved questions like the best division of a protein family into subfamilies, or the accurate detection of sequence variation patterns characteristic of different subfamilies. Here we present a systematic study in a significant number of protein families, testing the statistical meaning of the Tree-determinant residues predicted by three different methods that represent the range of available approaches. The first method takes as a starting point a phylogenetic representation of a protein family and, following the principle of Relative Entropy from Information Theory, automatically searches for the optimal division of the family into subfamilies. The second method looks for positions whose mutational behavior is reminiscent of the mutational behavior of the full-length proteins, by directly comparing the corresponding distance matrices. The third method is an automation of the analysis of distribution of sequences and amino acid positions in the corresponding multidimensional spaces using a vector-based principal component analysis. These three methods have been tested on two non-redundant lists of protein families: one composed by proteins that bind a variety of ligand groups, and the other composed by proteins with annotated functionally relevant sites. In most cases, the residues predicted by the three methods show a clear tendency to be close to bound ligands of biological relevance and to those amino acids described as participants in key aspects of protein function. These three automatic methods provide a wide range of possibilities for biologists to analyze their families of interest, in a similar way to the one presented here for the family of proteins related with ras-p21.     

Remote protein homology detection using recurrence quantification analysis and amino acid physicochemical properties

Remote homology detection refers to the detection of structure homology in evolutionarily related proteins with low sequence similarity. Supervised learning algorithms such as support vector machine (SVM) are currently the most accurate methods. In most of these SVM-based methods, efforts have been dedicated to developing new kernels to better use the pairwise alignment scores or sequence profiles. Moreover, amino acids’ physicochemical properties are not generally used in the feature representation of protein sequences. In this article, we present a remote homology detection method that incorporates two novel features: (1) a protein's primary sequence is represented using amino acid's physicochemical properties and (2) the similarity between two proteins is measured using recurrence quantification analysis (RQA). An optimization scheme was developed to select different amino acid indices (up to 10 for a protein family) that are best to characterize the given protein family. The selected amino acid indices may enable us to draw better biological explanation of the protein family classification problem than using other alignment-based methods. An SVM-based classifier will then work on the space described by the RQA metrics. The classification scheme is named as SVM-RQA. Experiments at the superfamily level of the SCOP1.53 dataset show that, without using alignment or sequence profile information, the features generated from amino acid indices are able to produce results that are comparable to those obtained by the published state-of-the-art SVM kernels. In the future, better prediction accuracies can be expected by combining the alignment-based features with our amino acids property-based features. Supplementary information including the raw dataset, the best-performing amino acid indices for each protein family and the computed RQA metrics for all protein sequences can be downloaded from http://ym151113.ym.edu.tw/svm-rqa.     

Reduced alphabet motif methodology for GPCR annotation

Identification and Classification of G-protein coupled receptors (GPCRs) using protein sequences is an important computational challenge, given that experimental screening of thousands of ligands is an expensive proposition. There are two distinct but complementary approaches to GPCR classification --machine learning and sequence motif analysis. Machine learning methodologies typically suffer from problems of class imbalance and lack of multi-class classification. Many sequence motif methods, meanwhile, are too dependent on the similarity of the primary sequence alignments. It is desirable to have a motif discovery and application methodology that is not strongly dependent on primary sequence similarity. It should also overcome limitations of machine learning. We propose and evaluate the effectiveness of a simple methodology that uses a reduced protein functional alphabet representation, where similar functional residues have similar symbols. Regular expression motifs can then be obtained by ClustalW based multiple sequence alignment, using an identity matrix. Since evolutionary matrices like BLOSUM, PAM are not used, this method can be useful for any set of sequences that do not necessarily share a common ancestry. Reduced alphabet motifs can accurately classify known GPCR proteins and the results are comparable to PRINTS and PROSITE. For well known GPCR proteins from SWISSPROT, there were no false negatives and only a few false positives. This methodology covers most currently known classes of GPCRs, even if there are very few representative sequences. It also predicts more than one class for certain sequences, thus overcoming the limitation of machine learning methods. We also annotated, 695 orphan receptors, and 121 were identified as belonging to Family A. A simple JavaScript based web interface has been developed to predict GPCR families and subfamilies (www.insilico-consulting.com/gpcrmotif.html).     

Characterization and prediction of lysine (K)-acetyl-transferase specific acetylation sites

Lysine acetylation is a well-studied post-translational modification on both histone and nonhistone proteins. More than 2000 acetylated proteins and 4000 lysine acetylation sites have been identified by large scale mass spectrometry or traditional experimental methods. Although over 20 lysine (K)-acetyl-transferases (KATs) have been characterized, which KAT is responsible for a given protein or lysine site acetylation is mostly unknown. In this work, we collected KAT-specific acetylation sites manually and analyzed sequence features surrounding the acetylated lysine of substrates from three main KAT families (CBP/p300, GCN5/PCAF, and the MYST family). We found that each of the three KAT families acetylates lysines with different sequence features. Based on these differences, we developed a computer program, Acetylation Set Enrichment Based method to predict which KAT-families are responsible for acetylation of a given protein or lysine site. Finally, we evaluated the efficiency of our method, and experimentally detected four proteins that were predicted to be acetylated by two KAT families when one representative member of the KAT family is over expressed. We conclude that our approach, combined with more traditional experimental methods, may be useful for identifying KAT families responsible for acetylated substrates proteome-wide.     

SVM-Prot: Web-based support vector machine software for functional classification of a protein from its primary sequence

Prediction of protein function is of significance in studying biological processes. One approach for function prediction is to classify a protein into functional family. Support vector machine (SVM) is a useful method for such classification, which may involve proteins with diverse sequence distribution. We have developed a web-based software, SVMProt, for SVM classification of a protein into functional family from its primary sequence. SVMProt classification system is trained from representative proteins of a number of functional families and seed proteins of Pfam curated protein families. It currently covers 54 functional families and additional families will be added in the near future. The computed accuracy for protein family classification is found to be in the range of 69.1-99.6%. SVMProt shows a certain degree of capability for the classification of distantly related proteins and homologous proteins of different function and thus may be used as a protein function prediction tool that complements sequence alignment methods. SVMProt can be accessed at http://jing.cz3.nus.edu.sg/cgi-bin/svmprot.cgi.     

Phylogenetic information improves homology detection.

We present a database search method that is based on phylogenetic trees (treesearch). The method is used to search a protein sequence database for homologs to a protein family. In preparation for the search, a phylogenetic tree is constructed from a given multiple alignment of the family. During the search, each database sequence is temporarily inserted into the tree, thus adding a new edge to the tree. Homology between family and sequence is then judged from the length of this edge. In a comparison of our method to profiles (ISREC pfsearch), two implementations of hidden Markov models (HMMER hmmsearch and SAM hmmscore), and to the family pairwise search (FPS) method on 43 families from the SCOP database based on minimum false-positive counts (min-FPCs), we found a considerable gain in sensitivity. In 69% of the test cases, treesearch showed a min-FPC of at most 50, whereas the two second best methods (hmmsearch and FPS) showed this performance only in 53% cases. A similar impression holds for a large range of min-FPC thresholds. The results demonstrate that phylogenetic information can significantly improve the detection of distant homologies and justify our method as a useful alternative to existing methods.     

Topological maps of protein sequences

A new method based on neural networks to cluster proteins into families is described. The network is trained with the Kohonen unsupervised learning algorithm, using matrix pattern representations of the protein sequences as inputs. The components (x, y) of these 20×20 matrix patterns are the normalized frequencies of all pairs xy of amino acids in each sequence. We investigate the influence of different learning parameters in the final topological maps obtained with a learning set of ten proteins belonging to three established families. In all cases, except in those where the synaptic vectors remains nearly unchanged during learning, the ten proteins are correctly classified into the expected families. The classification by the trained network of mutated or incomplete sequences of the learned proteins is also analysed. The neural network gives a correct classification for a sequence mutated in 21.5%±7% of its amino acids and for fragments representing 7.5%±3% of the original sequence. Similar results were obtained with a learning set of 32 proteins belonging to 15 families. These results show that a neural network can be trained following the Kohonen algorithm to obtain topological maps of protein sequences, where related proteins are finally associated to the same winner neuron or to neighboring ones, and that the trained network can be applied to rapidly classify new sequences. This approach opens new possibilities to find rapid and efficient algorithms to organize and search for homologies in the whole protein database.     

Optimal classification of protein sequences and selection of representative sets from multiple alignments: application to homologous families and lessons for structural genomics

Hierarchical classification is probably the most popular approach to group related proteins. However, there are a number of problems associated with its use for this purpose. One is that the resulting tree showing a nested sequence of groups may not be the most suitable representation of the data. Another is that visual inspection is the most common method to decide the most appropriate number of subsets from a tree. In fact, classification of proteins in general is bedevilled with the need for subjective thresholds to define group membership (e.g., 'significant' sequence identity for homologous families). Such arbitrariness is not only intellectually unsatisfying but also has important practical consequences. For instance, it hinders meaningful identification of protein targets for structural genomics. I describe an alternative approach to cluster related proteins without the need for an a priori threshold: one, through its use of dynamic programming, which is guaranteed to produce globally optimal solutions at all levels of partition granularity. Grouping proteins according to weights assigned to their aligned sequences makes it possible to delineate dynamically a 'core-periphery' structure within families. The 'core' of a protein family comprises the most typical sequences while the 'periphery' consists of the atypical ones. Further, a new sequence weighting scheme that combines the information in all the multiply aligned positions of an alignment in a novel way is put forward. Instead of averaging over all positions, this procedure takes into account directly the distribution of sequence variability along an alignment. The relationships between sequence weights and sequence identity are investigated for 168 families taken from HOMSTRAD, a database of protein structure alignments for homologous families. An exact solution is presented for the problem of how to select the most representative pair of sequences for a protein family. Extension of this approach by a greedy algorithm allows automatic identification of a minimal set of aligned sequences. The results of this analysis are available on the Web at http://mathbio.nimr.mrc.ac.uk/~amay.     

PALI-a database of Phylogeny and ALIgnment of homologous protein structures

PALI (release 1.2) contains three-dimensional (3-D) structure-dependent sequence alignments as well as structure-based phylogenetic trees of homologous protein domains in various families. The data set of homologous protein structures has been derived by consulting the SCOP database (release 1.50) and the data set comprises 604 families of homologous proteins involving 2739 protein domain structures with each family made up of at least two members. Each member in a family has been structurally aligned with every other member in the same family (pairwise alignment) and all the members in the family are also aligned using simultaneous super-position (multiple alignment). The structural alignments are performed largely automatically, with manual interventions especially in the cases of distantly related proteins, using the program STAMP (version 4.2). Every family is also associated with two dendrograms, calculated using PHYLIP (version 3.5), one based on a structural dissimilarity metric defined for every pairwise alignment and the other based on similarity of topologically equivalent residues. These dendrograms enable easy comparison of sequence and structure-based relationships among the members in a family. Structure-based alignments with the details of structural and sequence similarities, superposed coordinate sets and dendrograms can be accessed conveniently using a web interface. The database can be queried for protein pairs with sequence or structural similarities falling within a specified range. Thus PALI forms a useful resource to help in analysing the relationship between sequence and structure variation at a given level of sequence similarity. PALI also contains over 653 'orphans' (single member families). Using the web interface involving PSI_BLAST and PHYLIP it is possible to associate the sequence of a new protein with one of the families in PALI and generate a phylogenetic tree combining the query sequence and proteins of known 3-D structure. The database with the web interfaced search and dendrogram generation tools can be accessed at http://pauling.mbu.iisc.ernet. in/ approximately pali.     

A path from primary protein sequence to ligand recognition

A novel method to organize protein structural information based solely on sequence is presented. The method clusters proteins into families that correlate with the three-dimensional protein structure and the conformation of the bound ligands. This procedure was applied to nicotinamide adenine dinucleotide [NAD(P)]-utilizing enzymes to identify a total of 94 sequence families, 53 of which are structurally characterized. Each of the structurally characterized proteins within a sequence family correlates to a single protein fold and to a common bound conformation of NAD(P). A wide range of structural folds is identified that recognize NAD(P), including Rossmann folds and beta/alpha barrels. The defined sequence families can be used to identify the type and prevalence of NAD(P)-utilizing enzymes in the proteomes of sequenced organisms. The proteome of Mycobacterium tuberculosis was mined to generate a proteome-wide profile of NAD(P)-utilizing enzymes coded by this organism. This enzyme family comprises approximately 6% of the open reading frames, with the largest subgroup being the Rossmann fold, short-chain dehydrogenases. The preponderance of short-chain dehydrogenases correlates strongly with the phenotype of M. tuberculosis, which is characterized as having one of the most complex prokaryotic cell walls.     

A simple topological representation of protein structure: implications for new, fast, and robust structural classification

A topological representation of proteins is developed that makes use of two metrics: the Euclidean metric for identifying natural nearest neighboring residues via the Delaunay tessellation in Cartesian space and the distance between residues in sequence space. Using this representation, we introduce a quantitative and computationally inexpensive method for the comparison of protein structural topology. The method ultimately results in a numerical score quantifying the distance between proteins in a heuristically defined topological space. The properties of this scoring scheme are investigated and correlated with the standard Calpha distance root-mean-square deviation measure of protein similarity calculated by rigid body structural alignment. The topological comparison method is shown to have a characteristic dependence on protein conformational differences and secondary structure. This distinctive behavior is also observed in the comparison of proteins within families of structural relatives. The ability of the comparison method to successfully classify proteins into classes, superfamilies, folds, and families that are consistent with standard classification methods, both automated and human-driven, is demonstrated. Furthermore, it is shown that the scoring method allows for a fine-grained classification on the family, protein, and species level that agrees very well with currently established phylogenetic hierarchies. This fine classification is achieved without requiring visual inspection of proteins, sequence analysis, or the use of structural superimposition methods. Implications of the method for a fast, automated, topological hierarchical classification of proteins are discussed.