Theodore D. Liakopoulos

PRED-TMBB: a web server for predicting the topology of β-barrel outer membrane proteins

The beta-barrel outer membrane proteins constitute one of the two known structural classes of membrane proteins. Whereas there are several different web-based predictors for alpha-helical membrane proteins, currently there is no freely available prediction method for beta-barrel membrane proteins, at least with an acceptable level of accuracy. We present here a web server (PRED-TMBB, http://bioinformatics.biol.uoa.gr/PRED-TMBB) which is capable of predicting the transmembrane strands and the topology of beta-barrel outer membrane proteins of Gram-negative bacteria. The method is based on a Hidden Markov Model, trained according to the Conditional Maximum Likelihood criterion. The model was retrained and the training set now includes 16 non-homologous outer membrane proteins with structures known at atomic resolution. The user may submit one sequence at a time and has the option of choosing between three different decoding methods. The server reports the predicted topology of a given protein, a score indicating the probability of the protein being an outer membrane beta-barrel protein, posterior probabilities for the transmembrane strand prediction and a graphical representation of the assumed position of the transmembrane strands with respect to the lipid bilayer.

A Hidden Markov Model method, capable of predicting and discriminating β-barrel outer membrane proteins

BackgroundIntegral membrane proteins constitute about 20–30% of all proteins in the fully sequenced genomes. They come in two structural classes, the α-helical and the β-barrel membrane proteins, demonstrating different physicochemical characteristics, structure and localization. While transmembrane segment prediction for the α-helical integral membrane proteins appears to be an easy task nowadays, the same is much more difficult for the β-barrel membrane proteins. We developed a method, based on a Hidden Markov Model, capable of predicting the transmembrane β-strands of the outer membrane proteins of gram-negative bacteria, and discriminating those from water-soluble proteins in large datasets. The model is trained in a discriminative manner, aiming at maximizing the probability of correct predictions rather than the likelihood of the sequences.ResultsThe training has been performed on a non-redundant database of 14 outer membrane proteins with structures known at atomic resolution; it has been tested with a jacknife procedure, yielding a per residue accuracy of 84.2% and a correlation coefficient of 0.72, whereas for the self-consistency test the per residue accuracy was 88.1% and the correlation coefficient 0.824. The total number of correctly predicted topologies is 10 out of 14 in the self-consistency test, and 9 out of 14 in the jacknife. Furthermore, the model is capable of discriminating outer membrane from water-soluble proteins in large-scale applications, with a success rate of 88.8% and 89.2% for the correct classification of outer membrane and water-soluble proteins respectively, the highest rates obtained in the literature. That test has been performed independently on a set of known outer membrane proteins with low sequence identity with each other and also with the proteins of the training set.ConclusionBased on the above, we developed a strategy, that enabled us to screen the entire proteome of E. coli for outer membrane proteins. The results were satisfactory, thus the method presented here appears to be suitable for screening entire proteomes for the discovery of novel outer membrane proteins. A web interface available for non-commercial users is located at: http://bioinformatics.biol.uoa.gr/PRED-TMBB, and it is the only freely available HMM-based predictor for β-barrel outer membrane protein topology.

Combined prediction of Tat and Sec signal peptides with hidden Markov models

MOTIVATION Computational prediction of signal peptides is of great importance in computational biology. In addition to the general secretory pathway (Sec), Bacteria, Archaea and chloroplasts possess another major pathway that utilizes the Twin-Arginine translocase (Tat), which recognizes longer and less hydrophobic signal peptides carrying a distinctive pattern of two consecutive Arginines (RR) in the n-region. A major functional differentiation between the Sec and Tat export pathways lies in the fact that the former translocates secreted proteins unfolded through a protein-conducting channel, whereas the latter translocates completely folded proteins using an unknown mechanism. The purpose of this work is to develop a novel method for predicting and discriminating Sec from Tat signal peptides at better accuracy. RESULTS We report the development of a novel method, PRED-TAT, which is capable of discriminating Sec from Tat signal peptides and predicting their cleavage sites. The method is based on Hidden Markov Models and possesses a modular architecture suitable for both Sec and Tat signal peptides. On an independent test set of experimentally verified Tat signal peptides, PRED-TAT clearly outperforms the previously proposed methods TatP and TATFIND, whereas, when evaluated as a Sec signal peptide predictor compares favorably to top-scoring predictors such as SignalP and Phobius. The method is freely available for academic users at http://www.compgen.org/tools/PRED-TAT/.

Evaluation of methods for predicting the topology of β-barrel outer membrane proteins and a consensus prediction method

BackgroundPrediction of the transmembrane strands and topology of β-barrel outer membrane proteins is of interest in current bioinformatics research. Several methods have been applied so far for this task, utilizing different algorithmic techniques and a number of freely available predictors exist. The methods can be grossly divided to those based on Hidden Markov Models (HMMs), on Neural Networks (NNs) and on Support Vector Machines (SVMs). In this work, we compare the different available methods for topology prediction of β-barrel outer membrane proteins. We evaluate their performance on a non-redundant dataset of 20 β-barrel outer membrane proteins of gram-negative bacteria, with structures known at atomic resolution. Also, we describe, for the first time, an effective way to combine the individual predictors, at will, to a single consensus prediction method.ResultsWe assess the statistical significance of the performance of each prediction scheme and conclude that Hidden Markov Model based methods, HMM-B2TMR, ProfTMB and PRED-TMBB, are currently the best predictors, according to either the per-residue accuracy, the segments overlap measure (SOV) or the total number of proteins with correctly predicted topologies in the test set. Furthermore, we show that the available predictors perform better when only transmembrane β-barrel domains are used for prediction, rather than the precursor full-length sequences, even though the HMM-based predictors are not influenced significantly. The consensus prediction method performs significantly better than each individual available predictor, since it increases the accuracy up to 4% regarding SOV and up to 15% in correctly predicted topologies.ConclusionsThe consensus prediction method described in this work, optimizes the predicted topology with a dynamic programming algorithm and is implemented in a web-based application freely available to non-commercial users at http://bioinformatics.biol.uoa.gr/ConBBPRED.

Prediction of lipoprotein signal peptides in Gram-positive bacteria with a Hidden Markov Model.

We present a Hidden Markov Model method for the prediction of lipoprotein signal peptides of Gram-positive bacteria, trained on a set of 67 experimentally verified lipoproteins. The method outperforms LipoP and the methods based on regular expression patterns, in various data sets containing experimentally characterized lipoproteins, secretory proteins, proteins with an N-terminal TM segment and cytoplasmic proteins. The method is also very sensitive and specific in the detection of secretory signal peptides and in terms of overall accuracy outperforms even SignalP, which is the top-scoring method for the prediction of signal peptides. PRED-LIPO is freely available at http://bioinformatics.biol.uoa.gr/PRED-LIPO/, and we anticipate that it will be a valuable tool for the experimentalists studying secreted proteins and lipoproteins from Gram-positive bacteria.

Algorithms for incorporating prior topological information in HMMs: application to transmembrane proteins

BackgroundHidden Markov Models (HMMs) have been extensively used in computational molecular biology, for modelling protein and nucleic acid sequences. In many applications, such as transmembrane protein topology prediction, the incorporation of limited amount of information regarding the topology, arising from biochemical experiments, has been proved a very useful strategy that increased remarkably the performance of even the top-scoring methods. However, no clear and formal explanation of the algorithms that retains the probabilistic interpretation of the models has been presented so far in the literature.ResultsWe present here, a simple method that allows incorporation of prior topological information concerning the sequences at hand, while at the same time the HMMs retain their full probabilistic interpretation in terms of conditional probabilities. We present modifications to the standard Forward and Backward algorithms of HMMs and we also show explicitly, how reliable predictions may arise by these modifications, using all the algorithms currently available for decoding HMMs. A similar procedure may be used in the training procedure, aiming at optimizing the labels of the HMMs classes, especially in cases such as transmembrane proteins where the labels of the membrane-spanning segments are inherently misplaced. We present an application of this approach developing a method to predict the transmembrane regions of alpha-helical membrane proteins, trained on crystallographically solved data. We show that this method compares well against already established algorithms presented in the literature, and it is extremely useful in practical applications.ConclusionThe algorithms presented here, are easily implemented in any kind of a Hidden Markov Model, whereas the prediction method (HMM-TM) is freely available for academic users at http://bioinformatics.biol.uoa.gr/HMM-TM, offering the most advanced decoding options currently available.

Prediction of cell wall sorting signals in gram-positive bacteria with a hidden markov model: application to complete genomes.

Surface proteins in Gram-positive bacteria are frequently implicated in virulence. We have focused on a group of extracellular cell wall-attached proteins (CWPs), containing an LPXTG motif for cleavage and covalent coupling to peptidoglycan by sortase enzymes. A hidden Markov model (HMM) approach for predicting the LPXTG-anchored cell wall proteins of Gram-positive bacteria was developed and compared against existing methods. The HMM model is parsimonious in terms of the number of freely estimated parameters, and it has proved to be very sensitive and specific in a training set of 55 experimentally verified LPXTG-anchored cell wall proteins as well as in reliable data sets of globular and transmembrane proteins. In order to identify such proteins in Gram-positive bacteria, a comprehensive analysis of 94 completely sequenced genomes has been performed. We identified, in total, 860 LPXTG-anchored cell wall proteins, a number that is significantly higher compared to those obtained by other available methods. Of these proteins, 237 are hypothetical proteins according to the annotation of SwissProt, and 88 had no homologs in the SwissProt database--this might be evidence that they are members of newly identified families of CWPs. The prediction tool, the database with the proteins identified in the genomes, and supplementary material are available online at http://bioinformatics.biol.uoa.gr/CW-PRED/.

ExTopoDB: a database of experimentally derived topological models of transmembrane proteins

UNLABELLED ExTopoDB is a publicly accessible database of experimentally derived topological models of transmembrane proteins. It contains information collected from studies in the literature that report the use of biochemical methods for the determination of the topology of α-helical transmembrane proteins. Transmembrane protein topology is highly important in order to understand their function and ExTopoDB provides an up to date, complete and comprehensive dataset of experimentally determined topologies of α-helical transmembrane proteins. Topological information is combined with transmembrane topology prediction resulting in more reliable topological models. AVAILABILITY http://bioinformatics.biol.uoa.gr/ExTopoDB.

CW-PRED: a HMM-Based method for the classification of cell wall-anchored proteins of gram-positive bacteria

Gram-positive bacteria have surface proteins that are often implicated in virulence. A group of extracellular proteins attached to the cell wall contains an LPXTG-like motif that is target for cleavage and covalent coupling to peptidoglycan by sortase enzymes. A Hidden Markov Model (HMM) was developed for predicting the LPXTG and LPXTG-like cell-wall proteins of Gram-positive bacteria. The model is the first capable of predicting alternative (i.e. other than LPXTG-containing) substrates. Our analysis of 177 completely sequenced genomes identified 1456 cell-wall proteins, a number larger compared to the previously available methods. Among these, apart from the previously identified 1283 proteins carrying the LPXTG motif, we identified 39 newly identified proteins carrying NPXTG, 53 carrying LPXTA and 81 carrying the LAXTG motif. The tool is freely available for academic use at http://bioinformatics.biol.uoa.gr/CW-PRED/.

Faster Gradient Descent Training of Hidden Markov Models, Using Individual Learning Rate Adaptation

Hidden Markov Models (HMMs) are probabilistic models, suitable for a wide range of pattern recognition tasks. In this work, we propose a new gradient descent method for Conditional Maximum Likelihood (CML) training of HMMs, which significantly outperforms traditional gradient descent. Instead of using fixed learning rate for every adjustable parameter of the HMM, we propose the use of independent learning rate/step-size adaptation, which has been proved valuable as a strategy in Artificial Neural Networks training. We show here that our approach compared to standard gradient descent performs significantly better. The convergence speed is increased up to five times, while at the same time the training procedure becomes more robust, as tested on ap-plications from molecular biology. This is accomplished without additional computational complexity or the need for parameter tuning.