Mismatch string kernels for discriminative protein classification

MOTIVATION: Classification of proteins sequences into functional and structural families based on sequence homology is a central problem in computational biology. Discriminative supervised machine learning approaches provide good performance, but simplicity and computational efficiency of training and prediction are also important concerns. RESULTS: We introduce a class of string kernels, called mismatch kernels, for use with support vector machines (SVMs) in a discriminative approach to the problem of protein classification and remote homology detection. These kernels measure sequence similarity based on shared occurrences of fixed-length patterns in the data, allowing for mutations between patterns. Thus, the kernels provide a biologically well-motivated way to compare protein sequences without relying on family-based generative models such as hidden Markov models. We compute the kernels efficiently using a mismatch tree data structure, allowing us to calculate the contributions of all patterns occurring in the data in one pass while traversing the tree. When used with an SVM, the kernels enable fast prediction on test sequences. We report experiments on two benchmark SCOP datasets, where we show that the mismatch kernel used with an SVM classifier performs competitively with state-of-the-art methods for homology detection, particularly when very few training examples are available. Examination of the highest-weighted patterns learned by the SVM classifier recovers biologically important motifs in protein families and superfamilies.     

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.     

Searching Remote Homology with Spectral Clustering with Symmetry in Neighborhood Cluster Kernels

Remote homology detection among proteins utilizing only the unlabelled sequences is a central problem in comparative genomics. The existing cluster kernel methods based on neighborhoods and profiles and the Markov clustering algorithms are currently the most popular methods for protein family recognition. The deviation from random walks with inflation or dependency on hard threshold in similarity measure in those methods requires an enhancement for homology detection among multi-domain proteins. We propose to combine spectral clustering with neighborhood kernels in Markov similarity for enhancing sensitivity in detecting homology independent of “recent” paralogs. The spectral clustering approach with new combined local alignment kernels more effectively exploits the unsupervised protein sequences globally reducing inter-cluster walks. When combined with the corrections based on modified symmetry based proximity norm deemphasizing outliers, the technique proposed in this article outperforms other state-of-the-art cluster kernels among all twelve implemented kernels. The comparison with the state-of-the-art string and mismatch kernels also show the superior performance scores provided by the proposed kernels. Similar performance improvement also is found over an existing large dataset. Therefore the proposed spectral clustering framework over combined local alignment kernels with modified symmetry based correction achieves superior performance for unsupervised remote homolog detection even in multi-domain and promiscuous domain proteins from Genolevures database families with better biological relevance. Source code available upon request. Contact: rf.irbal@rakras.     

Profile-based direct kernels for remote homology detection and fold recognition

MOTIVATION: Protein remote homology detection is a central problem in computational biology. Supervised learning algorithms based on support vector machines are currently one of the most effective methods for remote homology detection. The performance of these methods depends on how the protein sequences are modeled and on the method used to compute the kernel function between them. RESULTS: We introduce two classes of kernel functions that are constructed by combining sequence profiles with new and existing approaches for determining the similarity between pairs of protein sequences. These kernels are constructed directly from these explicit protein similarity measures and employ effective profile-to-profile scoring schemes for measuring the similarity between pairs of proteins. Experiments with remote homology detection and fold recognition problems show that these kernels are capable of producing results that are substantially better than those produced by all of the existing state-of-the-art SVM-based methods. In addition, the experiments show that these kernels, even when used in the absence of profiles, produce results that are better than those produced by existing non-profile-based schemes. AVAILABILITY: The programs for computing the various kernel functions are available on request from the authors.     

Remote homolog detection using local sequence-structure correlations

Remote homology detection refers to the detection of structural homology in proteins when there is little or no sequence similarity. In this article, we present a remote homolog detection method called SVM-HMMSTR that overcomes the reliance on detectable sequence similarity by transforming the sequences into strings of hidden Markov states that represent local folding motif patterns. These state strings are transformed into fixed-dimension feature vectors for input to a support vector machine. Two sets of features are defined: an order-independent feature set that captures the amino acid and local structure composition; and an order-dependent feature set that captures the sequential ordering of the local structures. Tests using the Structural Classification of Proteins (SCOP) 1.53 data set show that the SVM-HMMSTR gives a significant improvement over several current methods.     

Remote homology detection: a motif based approach

MOTIVATION: Remote homology detection is the problem of detecting homology in cases of low sequence similarity. It is a hard computational problem with no approach that works well in all cases. RESULTS: We present a method for detecting remote homology that is based on the presence of discrete sequence motifs. The motif content of a pair of sequences is used to define a similarity that is used as a kernel for a Support Vector Machine (SVM) classifier. We test the method on two remote homology detection tasks: prediction of a previously unseen SCOP family and prediction of an enzyme class given other enzymes that have a similar function on other substrates. We find that it performs significantly better than an SVM method that uses BLAST or Smith-Waterman similarity scores as features.     

Profile-based string kernels for remote homology detection and motif extraction

We introduce novel profile-based string kernels for use with support vector machines (SVMs) for the problems of protein classification and remote homology detection. These kernels use probabilistic profiles, such as those produced by the PSI-BLAST algorithm, to define position-dependent mutation neighborhoods along protein sequences for inexact matching of k-length subsequences ("k-mers") in the data. By use of an efficient data structure, the kernels are fast to compute once the profiles have been obtained. For example, the time needed to run PSI-BLAST in order to build the profiles is significantly longer than both the kernel computation time and the SVM training time. We present remote homology detection experiments based on the SCOP database where we show that profile-based string kernels used with SVM classifiers strongly outperform all recently presented supervised SVM methods. We further examine how to incorporate predicted secondary structure information into the profile kernel to obtain a small but significant performance improvement. We also show how we can use the learned SVM classifier to extract "discriminative sequence motifs"--short regions of the original profile that contribute almost all the weight of the SVM classification score--and show that these discriminative motifs correspond to meaningful structural features in the protein data. The use of PSI-BLAST profiles can be seen as a semi-supervised learning technique, since PSI-BLAST leverages unlabeled data from a large sequence database to build more informative profiles. Recently presented "cluster kernels" give general semi-supervised methods for improving SVM protein classification performance. We show that our profile kernel results also outperform cluster kernels while providing much better scalability to large datasets.     

Stem kernels for RNA sequence analyses

Several computational methods based on stochastic context-free grammars have been developed for modeling and analyzing functional RNA sequences. These grammatical methods have succeeded in modeling typical secondary structures of RNA, and are used for structural alignment of RNA sequences. However, such stochastic models cannot sufficiently discriminate member sequences of an RNA family from nonmembers and hence detect noncoding RNA regions from genome sequences. A novel kernel function, stem kernel, for the discrimination and detection of functional RNA sequences using support vector machines (SVMs) is proposed. The stem kernel is a natural extension of the string kernel, specifically the all-subsequences kernel, and is tailored to measure the similarity of two RNA sequences from the viewpoint of secondary structures. The stem kernel examines all possible common base pairs and stem structures of arbitrary lengths, including pseudoknots between two RNA sequences, and calculates the inner product of common stem structure counts. An efficient algorithm is developed to calculate the stem kernels based on dynamic programming. The stem kernels are then applied to discriminate members of an RNA family from nonmembers using SVMs. The study indicates that the discrimination ability of the stem kernel is strong compared with conventional methods. Furthermore, the potential application of the stem kernel is demonstrated by the detection of remotely homologous RNA families in terms of secondary structures. This is because the string kernel is proven to work for the remote homology detection of protein sequences. These experimental results have convinced us to apply the stem kernel in order to find novel RNA families from genome sequences.     

Fast model-based protein homology detection without alignment

MOTIVATION: As more genomes are sequenced, the demand for fast gene classification techniques is increasing. To analyze a newly sequenced genome, first the genes are identified and translated into amino acid sequences which are then classified into structural or functional classes. The best-performing protein classification methods are based on protein homology detection using sequence alignment methods. Alignment methods have recently been enhanced by discriminative methods like support vector machines (SVMs) as well as by position-specific scoring matrices (PSSM) as obtained from PSI-BLAST. However, alignment methods are time consuming if a new sequence must be compared to many known sequences-the same holds for SVMs. Even more time consuming is to construct a PSSM for the new sequence. The best-performing methods would take about 25 days on present-day computers to classify the sequences of a new genome (20,000 genes) as belonging to just one specific class--however, there are hundreds of classes. Another shortcoming of alignment algorithms is that they do not build a model of the positive class but measure the mutual distance between sequences or profiles. Only multiple alignments and hidden Markov models are popular classification methods which build a model of the positive class but they show low classification performance. The advantage of a model is that it can be analyzed for chemical properties common to the class members to obtain new insights into protein function and structure. We propose a fast model-based recurrent neural network for protein homology detection, the 'Long Short-Term Memory' (LSTM). LSTM automatically extracts indicative patterns for the positive class, but in contrast to profile methods it also extracts negative patterns and uses correlations between all detected patterns for classification. LSTM is capable to automatically extract useful local and global sequence statistics like hydrophobicity, polarity, volume, polarizability and combine them with a pattern. These properties make LSTM complementary to alignment-based approaches as it does not use predefined similarity measures like BLOSUM or PAM matrices. RESULTS: We have applied LSTM to a well known benchmark for remote protein homology detection, where a protein must be classified as belonging to a SCOP superfamily. LSTM reaches state-of-the-art classification performance but is considerably faster for classification than other approaches with comparable classification performance. LSTM is five orders of magnitude faster than methods which perform slightly better in classification and two orders of magnitude faster than the fastest SVM-based approaches (which, however, have lower classification performance than LSTM). Only PSI-BLAST and HMM-based methods show comparable time complexity as LSTM, but they cannot compete with LSTM in classification performance. To test the modeling capabilities of LSTM, we applied LSTM to PROSITE classes and interpreted the extracted patterns. In 8 out of 15 classes, LSTM automatically extracted the PROSITE motif. In the remaining 7 cases alternative motifs are generated which give better classification results on average than the PROSITE motifs. AVAILABILITY: The LSTM algorithm is available from http://www.bioinf.jku.at/software/LSTM_protein/.     

Iterative sequence/secondary structure search for protein homologs: comparison with amino acid sequence alignments and application to fold recognition in genome databases

MOTIVATION: Sequence alignment techniques have been developed into extremely powerful tools for identifying the folding families and function of proteins in newly sequenced genomes. For a sufficiently low sequence identity it is necessary to incorporate additional structural information to positively detect homologous proteins. We have carried out an extensive analysis of the effectiveness of incorporating secondary structure information directly into the alignments for fold recognition and identification of distant protein homologs. A secondary structure similarity matrix based on a database of three-dimensionally aligned proteins was first constructed. An iterative application of dynamic programming was used which incorporates linear combinations of amino acid and secondary structure sequence similarity scores. Initially, only primary sequence information is used. Subsequently contributions from secondary structure are phased in and new homologous proteins are positively identified if their scores are consistent with the predetermined error rate. RESULTS: We used the SCOP40 database, where only PDB sequences that have 40% homology or less are included, to calibrate homology detection by the combined amino acid and secondary structure sequence alignments. Combining predicted secondary structure with sequence information results in a 8-15% increase in homology detection within SCOP40 relative to the pairwise alignments using only amino acid sequence data at an error rate of 0.01 errors per query; a 35% increase is observed when the actual secondary structure sequences are used. Incorporating predicted secondary structure information in the analysis of six small genomes yields an improvement in the homology detection of approximately 20% over SSEARCH pairwise alignments, but no improvement in the total number of homologs detected over PSI-BLAST, at an error rate of 0.01 errors per query. However, because the pairwise alignments based on combinations of amino acid and secondary structure similarity are different from those produced by PSI-BLAST and the error rates can be calibrated, it is possible to combine the results of both searches. An additional 25% relative improvement in the number of genes identified at an error rate of 0.01 is observed when the data is pooled in this way. Similarly for the SCOP40 dataset, PSI-BLAST detected 15% of all possible homologs, whereas the pooled results increased the total number of homologs detected to 19%. These results are compared with recent reports of homology detection using sequence profiling methods. AVAILABILITY: Secondary structure alignment homepage at http://lutece.rutgers.edu/ssas CONTACT: anders@rutchem.rutgers.edu; ronlevy@lutece.rutgers.edu Supplementary Information: Genome sequence/structure alignment results at http://lutece.rutgers.edu/ss_fold_predictions.     

Physicochemical property distributions for accurate and rapid pairwise protein homology detection

Background  

The challenge of remote homology detection is that many evolutionarily related sequences have very little similarity at the amino acid level. Kernel-based discriminative methods, such as support vector machines (SVMs), that use vector representations of sequences derived from sequence properties have been shown to have superior accuracy when compared to traditional approaches for the task of remote homology detection.     

Genome-wide searching with base-pairing kernel functions for noncoding RNAs: computational and expression analysis of snoRNA families in Caenorhabditis elegans

Despite the accumulating research on noncoding RNAs (ncRNAs), it is likely that we are seeing only the tip of the iceberg regarding our understanding of the functions and the regulatory roles served by ncRNAs in cellular metabolism, pathogenesis and host-pathogen interactions. Therefore, more powerful computational and experimental tools for analyzing ncRNAs need to be developed. To this end, we propose novel kernel functions, called base-pairing profile local alignment (BPLA) kernels, for analyzing functional ncRNA sequences using support vector machines (SVMs). We extend the local alignment kernels for amino acid sequences in order to handle RNA sequences by using STRAL's; scoring function, which takes into account sequence similarities as well as upstream and downstream base-pairing probabilities, thus enabling us to model secondary structures of RNA sequences. As a test of the performance of BPLA kernels, we applied our kernels to the problem of discriminating members of an RNA family from nonmembers using SVMs. The results indicated that the discrimination ability of our kernels is stronger than that of other existing methods. Furthermore, we demonstrated the applicability of our kernels to the problem of genome-wide search of snoRNA families in the Caenorhabditis elegans genome, and confirmed that the expression is valid in 14 out of 48 of our predicted candidates by using qRT-PCR. Finally, highly expressed six candidates were identified as the original target regions by DNA sequencing.     

A discriminative method for remote homology detection based on n-peptide compositions with reduced amino acid alphabets

In this study, n-peptide compositions are utilized for protein vectorization over a discriminative remote homology detection framework based on support vector machines (SVMs). The size of amino acid alphabet is gradually reduced for increasing values of n to make the method to conform with the memory resources in conventional workstations. A hash structure is implemented for accelerated search of n-peptides. The method is tested to see its ability to classify proteins into families on a subset of SCOP family database and compared against many of the existing homology detection methods including the most popular generative methods; SAM-98 and PSI-BLAST and the recent SVM methods; SVM-Fisher, SVM-BLAST and SVM-Pairwise. The results have demonstrated that the new method significantly outperforms SVM-Fisher, SVM-BLAST, SAM-98 and PSI-BLAST, while achieving a comparable accuracy with SVM-Pairwise. In terms of efficiency, it performs much better than SVM-Pairwise. It is shown that the information of n-peptide compositions with reduced amino acid alphabets provides an accurate and efficient means of protein vectorization for SVM-based sequence classification.     

Multi-class protein fold recognition using support vector machines and neural networks

MOTIVATION: Protein fold recognition is an important approach to structure discovery without relying on sequence similarity. We study this approach with new multi-class classification methods and examined many issues important for a practical recognition system. RESULTS: Most current discriminative methods for protein fold prediction use the one-against-others method, which has the well-known 'False Positives' problem. We investigated two new methods: the unique one-against-others and the all-against-all methods. Both improve prediction accuracy by 14-110% on a dataset containing 27 SCOP folds. We used the Support Vector Machine (SVM) and the Neural Network (NN) learning methods as base classifiers. SVMs converges fast and leads to high accuracy. When scores of multiple parameter datasets are combined, majority voting reduces noise and increases recognition accuracy. We examined many issues involved with large number of classes, including dependencies of prediction accuracy on the number of folds and on the number of representatives in a fold. Overall, recognition systems achieve 56% fold prediction accuracy on a protein test dataset, where most of the proteins have below 25% sequence identity with the proteins used in training.     

Improved profile HMM performance by assessment of critical algorithmic features in SAM and HMMER

Background  

Profile hidden Markov model (HMM) techniques are among the most powerful methods for protein homology detection. Yet, the critical features for successful modelling are not fully known. In the present work we approached this by using two of the most popular HMM packages: SAM and HMMER. The programs' abilities to build models and score sequences were compared on a SCOP/Pfam based test set. The comparison was done separately for local and global HMM scoring.     

Application of String Kernels in Protein Sequence Classification

INTRODUCTION: The production of biological information has become much greater than its consumption. The key issue now is how to organise and manage the huge amount of novel information to facilitate access to this useful and important biological information. One core problem in classifying biological information is the annotation of new protein sequences with structural and functional features. METHOD: This article introduces the application of string kernels in classifying protein sequences into homogeneous families. A string kernel approach used in conjunction with support vector machines has been shown to achieve good performance in text categorisation tasks. We evaluated and analysed the performance of this approach, and we present experimental results on three selected families from the SCOP (Structural Classification of Proteins) database. We then compared the overall performance of this method with the existing protein classification methods on benchmark SCOP datasets. RESULTS: According to the F1 performance measure and the rate of false positive (RFP) measure, the string kernel method performs well in classifying protein sequences. The method outperformed all the generative-based methods and is comparable with the SVM-Fisher method. DISCUSSION: Although the string kernel approach makes no use of prior biological knowledge, it still captures sufficient biological information to enable it to outperform some of the state-of-the-art methods.     

Motif kernel generated by genetic programming improves remote homology and fold detection

Background  

Protein remote homology detection is a central problem in computational biology. Most recent methods train support vector machines to discriminate between related and unrelated sequences and these studies have introduced several types of kernels. One successful approach is to base a kernel on shared occurrences of discrete sequence motifs. Still, many protein sequences fail to be classified correctly for a lack of a suitable set of motifs for these sequences.     

A class of edit kernels for SVMs to predict translation initiation sites in eukaryotic mRNAs.

The prediction of translation initiation sites (TISs) in eukaryotic mRNAs has been a challenging problem in computational molecular biology. In this paper, we present a new algorithm to recognize TISs with a very high accuracy. Our algorithm includes two novel ideas. First, we introduce a class of new sequence-similarity kernels based on string editing, called edit kernels, for use with support vector machines (SVMs) in a discriminative approach to predict TISs. The edit kernels are simple and have significant biological and probabilistic interpretations. Although the edit kernels are not positive definite, it is easy to make the kernel matrix positive definite by adjusting the parameters. Second, we convert the region of an input mRNA sequence downstream to a putative TIS into an amino acid sequence before applying SVMs to avoid the high redundancy in the genetic code. The algorithm has been implemented and tested on previously published data. Our experimental results on real mRNA data show that both ideas improve the prediction accuracy greatly and that our method performs significantly better than those based on neural networks and SVMs with polynomial kernels or Salzberg kernels.