Generalized Baum-Welch Algorithm Based on the Similarity between Sequences

BackgroundIntrinsically Disordered Proteins (IDPs) lack an ordered three-dimensional structure and are enriched in various biological processes. The Molecular Recognition Features (MoRFs) are functional regions within IDPs that undergo a disorder-to-order transition on binding to a partner protein. Identifying MoRFs in IDPs using computational methods is a challenging task.MethodsIn this study, we introduce hidden Markov model (HMM) profiles to accurately identify the location of MoRFs in disordered protein sequences. Using windowing technique, HMM profiles are utilised to extract features from protein sequences and support vector machines (SVM) are used to calculate a propensity score for each residue. Two different SVM kernels with high noise tolerance are evaluated with a varying window size and the scores of the SVM models are combined to generate the final propensity score to predict MoRF residues. The SVM models are designed to extract maximal information between MoRF residues, its neighboring regions (Flanks) and the remainder of the sequence (Others).ResultsTo evaluate the proposed method, its performance was compared to that of other MoRF predictors; MoRFpred and ANCHOR. The results show that the proposed method outperforms these two predictors.ConclusionsUsing HMM profile as a source of feature extraction, the proposed method indicates improvement in predicting MoRFs in disordered protein sequences.

BackgroundProfile Hidden Markov Model (HMM) is a powerful statistical model to represent a family of DNA, RNA, and protein sequences. Profile HMM has been widely used in bioinformatics research such as sequence alignment, gene structure prediction, motif identification, protein structure prediction, and biological database search. However, few comprehensive, visual editing tools for profile HMM are publicly available.ResultsWe develop a visual editor for profile Hidden Markov Models (HMMEditor). HMMEditor can visualize the profile HMM architecture, transition probabilities, and emission probabilities. Moreover, it provides functions to edit and save HMM and parameters. Furthermore, HMMEditor allows users to align a sequence against the profile HMM and to visualize the corresponding Viterbi path.ConclusionHMMEditor provides a set of unique functions to visualize and edit a profile HMM. It is a useful tool for biological sequence analysis and modeling. Both HMMEditor software and web service are freely available.

Accurate prediction of protein structure relies heavily on exploiting multiple sequence alignment (MSA) for residue mutations and correlations as this information specifies protein tertiary structure. The widely used prediction approaches usually transform MSA into inter-mediate models, say position-specific scoring matrix or profile hidden Markov model. These inter-mediate models, however, cannot fully represent residue mutations and correlations carried by MSA; hence, an effective way to directly exploit MSAs is highly desirable. Here, we report a novel sequence set network (called Seq-SetNet) to directly and effectively exploit MSA for protein structure prediction. Seq-SetNet uses an 'encoding and aggregation' strategy that consists of two key elements: (i) an encoding module that takes a component homologue in MSA as input, and encodes residue mutations and correlations into context-specific features for each residue; and (ii) an aggregation module to aggregate the features extracted from all component homologues, which are further transformed into structural properties for residues of the query protein. As Seq-SetNet encodes each homologue protein individually, it could consider both insertions and deletions, as well as long-distance correlations among residues, thus representing more information than the inter-mediate models. Moreover, the encoding module automatically learns effective features and thus avoids manual feature engineering. Using symmetric aggregation functions, Seq-SetNet processes the homologue proteins as a sequence set, making its prediction results invariable to the order of these proteins. On popular benchmark sets, we demonstrated the successful application of Seq-SetNet to predict secondary structure and torsion angles of residues with improved accuracy and efficiency. The code and datasets are available through https://github.com/fusong-ju/Seq-SetNet. Supplementary data are available at Bioinformatics online.

This book explores important aspects of Markov and hidden Markov processes and the applications of these ideas to various problems in computational biology. It starts from first principles, so that no previous knowledge of probability is necessary. However, the work is rigorous and mathematical, making it useful to engineers and mathematicians, even those not interested in biological applications. A range of exercises is provided, including drills to familiarize the reader with concepts and more advanced problems that require deep thinking about the theory. Biological applications are taken from post-genomic biology, especially genomics and proteomics. The topics examined include standard material such as the Perron–Frobenius theorem, transient and recurrent states, hitting probabilities and hitting times, maximum likelihood estimation, the Viterbi algorithm, and the Baum–Welch algorithm. The book contains discussions of extremely useful topics not usually seen at the basic level, such as ergodicity of Markov processes, Markov Chain Monte Carlo (MCMC), information theory, and large deviation theory for both i.i.d and Markov processes. It also presents state-of-the-art realization theory for hidden Markov models. Among biological applications, it offers an in-depth look at the BLAST (Basic Local Alignment Search Technique) algorithm, including a comprehensive explanation of the underlying theory. Other applications such as profile hidden Markov models are also explored.

Acyl-coenzyme A synthetases (ACSs) catalyze the fundamental, initial reaction in fatty acid metabolism. "Activation" of fatty acids by thioesterification to CoA allows their participation in both anabolic and catabolic pathways. The availability of the sequenced human genome has facilitated the investigation of the number of ACS genes present. Using two conserved amino acid sequence motifs to probe human DNA databases, 26 ACS family genes/proteins were identified. ACS activity in either humans or rodents was demonstrated previously for 20 proteins, but 6 remain candidate ACSs. For two candidates, cDNA was cloned, protein was expressed in COS-1 cells, and ACS activity was detected. Amino acid sequence similarities were used to assign enzymes into subfamilies, and subfamily assignments were consistent with acyl chain length preference. Four of the 26 proteins did not fit into a subfamily, and bootstrap analysis of phylograms was consistent with evolutionary divergence. Three additional conserved amino acid sequence motifs were identified that likely have functional or structural roles. The existence of many ACSs suggests that each plays a unique role, directing the acyl-CoA product to a specific metabolic fate. Knowing the full complement of ACS genes in the human genome will facilitate future studies to characterize their specific biological functions.

Co-evolution has an important function in the evolution of species and it is clearly manifested in certain scenarios such as host–parasite and predator–prey interactions, symbiosis and mutualism. The extrapolation of the concepts and methodologies developed for the study of species co-evolution at the molecular level has prompted the development of a variety of computational methods able to predict protein interactions through the characteristics of co-evolution. Particularly successful have been those methods that predict interactions at the genomic level based on the detection of pairs of protein families with similar evolutionary histories (similarity of phylogenetic trees: mirrortree). Future advances in this field will require a better understanding of the molecular basis of the co-evolution of protein families. Thus, it will be important to decipher the molecular mechanisms underlying the similarity observed in phylogenetic trees of interacting proteins, distinguishing direct specific molecular interactions from other general functional constraints. In particular, it will be important to separate the effects of physical interactions within protein complexes (‘co-adaptation') from other forces that, in a less specific way, can also create general patterns of co-evolution.

A group of 16-kDa proteins, synthesized and secreted by rat pancreatic acinar cells and composed of pancreatic stone protein (PSP/reg) and isoforms of pancreatitis-associated protein (PAP), show structural homologies, including conserved amino acid sequences, cysteine residues, and highly sensitive N-terminal trypsin cleavage sites, as well as conserved functional responses in conditions of pancreatic stress. Trypsin activation of recombinant stress proteins or counterparts contained in rat pancreatic juice (PSP/reg, PAP I and PAP III) resulted in conversion of 16-kDa soluble proteins into 14-kDa soluble isoforms (pancreatic thread protein and pancreatitis-associated thread protein, respectively) that rapidly polymerize into insoluble sedimenting structures. Activated thread proteins show long lived resistance to a wide spectrum of proteases contained in pancreatic juice, including serine proteases and metalloproteinases. In contrast, PAP II, following activation with trypsin or pancreatic juice, does not form insoluble structures and is rapidly digested by pancreatic proteases. Scanning and transmission electron microscopy indicate that activated thread proteins polymerize into highly organized fibrillar structures with helical configurations. Through bundling, branching, and extension processes, these fibrillar structures form dense matrices that span large topological surfaces. These findings suggest that PSP/reg and PAP I and III isoforms consist of a family of highly regulated soluble secretory stress proteins, which, upon trypsin activation, convert into a family of insoluble helical thread proteins. Dense extracellular matrices, composed of helical thread proteins organized into higher ordered matrix structures, may serve physiological functions within luminal compartments in the exocrine pancreas.

Recognizing human actions from image sequences is an active area of research in computer vision. In this paper, a novel HMM-based approach is proposed for human action recognition using 3D positions of body joints. First, actions are segmented into meaningful action units called dynamic instants and intervals by using motion velocities, the direction of motion, and the curvatures of 3D trajectories. Then action unit with its spatio-temporal feature sets are clustered using unsupervised learning, like SOM, to generate a sequence of discrete symbols. To overcome an abrupt change or an abnormal in its gesticulation between different performances of the same action, Profile Hidden Markov Models (Profile HMMs) are applied with these symbol sequences using Viterbi and Baum-Welch algorithms for human activity recognition. The experimental evaluations show that the proposed approach achieves promising results compared to other state of the art algorithms.KeywordsView-invariant representationSkeleton jointsHuman activity recognitionProfile HMMSelf-organizing map

This paper shows how the newly developed paradigm of General Fuzzy Automata (GFA) can be used as a biological sequence predictor. We consider the positional correlations of amino acids in a protein family as the basic criteria for prediction and classification of unknown sequences. It will be shown how the GFA formalism can be used as an efficient tool for classification of protein sequences. The results show that this approach predicts the membership of an unknown sequence in a protein family better than profile Hidden Markov Models (HMMs) which are now a popular and putative approach in biological sequence analysis.

In this paper we propose a new framework for modeling 2D shapes. A shape is first described by a sequence of local features (e.g., curvature) of the shape boundary. The resulting description is then used to build a profile hidden Markov model (PHMM) representation of the shape. PHMMs are a particular type of hidden Markov models (HMMs) with special states and architecture that can tolerate considerable shape contour perturbations, including rigid and non-rigid deformations, occlusions and missing contour parts. Different from traditional HMM-based shape models, the sparseness of the PHMM structure allows efficient inference and learning algorithms for shape modeling and analysis. The new framework can be applied to a wide range of problems, from shape matching and classification to shape segmentation. Our experimental results show the effectiveness and robustness of this new approach in the three application domains.

InterPro (http://www.ebi.ac.uk/interpro/) is a freely available database used to classify protein sequences into families and to predict the presence of important domains and sites. The resource is based around diagnostic models (profile hidden Markov models, profiles, position-specific scoring matrices and regular expressions) provided by 14 different member databases, against which protein sequences are searched to determine their potential functions. Two recently added member databases, CDD and SLFD, also provide specific residue-level annotation. Their addition has enabled identification of important amino acids, such as those responsible for ligand-binding, protein-protein interactions, or comprising active sites, providing an additional tier to the annotations already provided by InterPro. In a further expansion, InterPro has also added prediction of intrinsically disordered regions (IDRs) – polypeptide segments that have little or no three-dimensional structure. IDRs have a wide range of potential functions, from acting as flexible linkers between domains, to interacting with other proteins and modifying their activity. Due to their bias composition and differing pattern of amino acid conservation, IDRs are hard to model using profile approaches, hence annotation of these regions has been limited to date. To address this, InterPro has integrated the MobiDB Lite tool, which combines eight different prediction methods to generate consensus IDR annotations, focusing on long disordered regions (>20 amino acids). This approach has yielded IDR annotations for ~ 24% of sequences in UniProtKB, covering ~ 5.5% of amino acid residues. Here we describe how the new per-residue and IDR developments in InterPro enable more richly-detailed and informative annotation of protein sequences, and how they can be used to support more specific functional inferences.

Profile hidden Markov models (profile HMMs) and probabilistic inference methods have made important contributions to the theory of sequence database homology search. However, practical use of profile HMM methods has been hindered by the computational expense of existing software implementations. Here I describe an acceleration heuristic for profile HMMs, the “multiple segment Viterbi” (MSV) algorithm. The MSV algorithm computes an optimal sum of multiple ungapped local alignment segments using a striped vector-parallel approach previously described for fast Smith/Waterman alignment. MSV scores follow the same statistical distribution as gapped optimal local alignment scores, allowing rapid evaluation of significance of an MSV score and thus facilitating its use as a heuristic filter. I also describe a 20-fold acceleration of the standard profile HMM Forward/Backward algorithms using a method I call “sparse rescaling”. These methods are assembled in a pipeline in which high-scoring MSV hits are passed on for reanalysis with the full HMM Forward/Backward algorithm. This accelerated pipeline is implemented in the freely available HMMER3 software package. Performance benchmarks show that the use of the heuristic MSV filter sacrifices negligible sensitivity compared to unaccelerated profile HMM searches. HMMER3 is substantially more sensitive and 100- to 1000-fold faster than HMMER2. HMMER3 is now about as fast as BLAST for protein searches.

While pairwise sequence alignment (PSA) by dynamic programming is guaranteed to generate one of the optimal alignments, multiple sequence alignment (MSA) of highly divergent sequences often results in poorly aligned sequences, plaguing all subsequent phylogenetic analysis. One way to avoid this problem is to use only PSA to reconstruct phylogenetic trees, which can only be done with distance-based methods. I compared the accuracy of this new computational approach (named PhyPA for phylogenetics by pairwise alignment) against the maximum likelihood method using MSA (the ML+MSA approach), based on nucleotide, amino acid and codon sequences simulated with different topologies and tree lengths. I present a surprising discovery that the fast PhyPA method consistently outperforms the slow ML+MSA approach for highly diverged sequences even when all optimization options were turned on for the ML+MSA approach. Only when sequences are not highly diverged (i.e., when a reliable MSA can be obtained) does the ML+MSA approach outperforms PhyPA. The true topologies are always recovered by ML with the true alignment from the simulation. However, with MSA derived from alignment programs such as MAFFT or MUSCLE, the recovered topology consistently has higher likelihood than that for the true topology. Thus, the failure to recover the true topology by the ML+MSA is not because of insufficient search of tree space, but by the distortion of phylogenetic signal by MSA methods. I have implemented in DAMBE PhyPA and two approaches making use of multi-gene data sets to derive phylogenetic support for subtrees equivalent to resampling techniques such as bootstrapping and jackknifing.

Embryos and larvae of sea urchins (Lytechinus variegatus, Strongylocentrotus droebachiensis, Strongylocentrotus purpuratus, Dendraster excentricus), and starfish (Pisaster ochraceus) were investigated for the presence of a functional endocannabinoid system. Anandamide (arachidonoyl ethanolamide, AEA), was measured in early L. variegatus embryos by liquid chromatography/mass spectrometry. AEA showed a strong developmental dynamic, increasing more than 5-fold between the 8–16 cell and mid-blastula 2 stage. ‘Perturb-and-rescue’ experiments in different sea urchin species and starfish showed that AEA blocked transition of embryos from the blastula to the gastrula stage, but had no effect on cleavage divisions, even at high doses. The non-selective cannabinoid receptor agonist, CP55940, had similar effects, but unlike AEA, also blocked cleavage divisions. CB1 antagonists, AEA transport inhibitors, and the cation channel transient membrane potential receptor V1 (TrpV1) agonist, arachidonoyl vanillic acid (arvanil), as well as arachidonoyl serotonin and dopamine (AA-5-HT, AA-DA) acted as rescue substances, partially or totally preventing abnormal embryonic phenotypes elicited by AEA or CP55940. Radioligand binding of [³H]CP55940 to membrane preparations from embryos/larvae failed to show significant binding, consistent with the lack of CB receptor orthologs in the sea urchin genome. However, when binding was conducted on whole cell lysates, a small amount of [³H]CP55940 binding was observed at the pluteus stage that was displaced by the CB2 antagonist, SR144528. Since AEA is known to bind with high affinity to TrpV1 and to certain G-protein-coupled receptors (GPCRs), the ability of arvanil, AA-5-HT and AA-DA to rescue embryos from AEA teratogenesis suggests that in sea urchins AEA and other endocannabinoids may utilize both Trp and GPCR orthologs. This possibility was explored using bioinformatic and phylogenetic tools to identify candidate orthologs in the S. purpuratus sea urchin genome. Candidate TrpA1 and TrpV1 orthologs were identified. The TrpA1 ortholog fell within a monophyletic clade, including both vertebrate and invertebrate orthologs, whereas the TrpV1 orthologs fell within two distinct TrpV-like invertebrate clades. One of the sea urchin TrpV orthologs was more closely related to the vertebrate epithelial calcium channels (TrpV5-6 family) than to the vertebrate TrpV1-4 family, as determined using profile-hidden Markov model (HMM) searches. Candidate dopamine and adrenergic GPCR orthologs were identified in the sea urchin genome, but no cannabinoid GPCRs were found, consistent with earlier studies. Candidate dopamine D₁, D₂ or α₁-adrenergic receptor orthologs were identified as potential progenitors to the vertebrate cannabinoid receptors using HMM searches, depending on whether the multiple sequence alignment of CB receptor sequences consisted only of urochordate and cephalochordate sequences or also included vertebrate sequences.

Hidden Markov Models are a widely used generative model for analysing sequence data. A variant, Profile Hidden Markov Models are a special case used in Bioinformatics to represent, for example, protein families. In this paper we introduce a simple propositionalisation method for Profile Hidden Markov Models. The method allows the use of PHMMs discriminatively in a classification task. Previously, kernel approaches have been proposed to generate a discriminative description for an HMM, but require the explicit definition of a similarity measure for HMMs. Propositionalisation does not need such a measure and allows the use of any propositional learner including kernel-based approaches. We show empirically that using propositionalisation leads to higher accuracies in comparison with PHMMs on benchmark datasets.