The influence of gapped positions in multiple sequence alignments on secondary structure prediction methods

All currently leading protein secondary structure prediction methods use a multiple protein sequence alignment to predict the secondary structure of the top sequence. In most of these methods, prior to prediction, alignment positions showing a gap in the top sequence are deleted, consequently leading to shrinking of the alignment and loss of position-specific information. In this paper we investigate the effect of this removal of information on secondary structure prediction accuracy. To this end, we have designed SymSSP, an algorithm that post-processes the predicted secondary structure of all sequences in a multiple sequence alignment by (i) making use of the alignment's evolutionary information and (ii) re-introducing most of the information that would otherwise be lost. The post-processed information is then given to a new dynamic programming routine that produces an optimally segmented consensus secondary structure for each of the multiple alignment sequences. We have tested our method on the state-of-the-art secondary structure prediction methods PHD, PROFsec, SSPro2 and JNET using the HOMSTRAD database of reference alignments. Our consensus-deriving dynamic programming strategy is consistently better at improving the segmentation quality of the predictions compared to the commonly used majority voting technique. In addition, we have applied several weighting schemes from the literature to our novel consensus-deriving dynamic programming routine. Finally, we have investigated the level of noise introduced by prediction errors into the consensus and show that predictions of edges of helices and strands are half the time wrong for all the four tested prediction methods.     

Computational methods for protein secondary structure prediction using multiple sequence alignments

Efforts to use computers in predicting the secondary structure of proteins based only on primary structure information started over a quarter century ago [1-3]. Although the results were encouraging initially, the accuracy of the pioneering methods generally did not attain the level required for using predictions of secondary structures reliably in modelling the three-dimensional topology of proteins. During the last decade, however, the introduction of new computational techniques as well as the use of multiple sequence information has lead to a dramatic increase in the success rate of prediction methods, such that successful 3D modelling based on predicted secondary structure has become feasible [e.g., Ref 4]. This review is aimed at presenting an overview of the scale of the secondary structure prediction problem and associated pitfalls, as well as the history of the development of computational prediction methods. As recent successful strategies for secondary structure prediction all rely on multiple sequence information, some methods for accurate protein multiple sequence alignments will also be described. While the main focus is on prediction methods for globular proteins, also the prediction of trans-membrane segments within membrane proteins will be briefly summarised. Finally, an integrated iterative approach tying secondary structure prediction and multiple alignment will be introduced [5].     

A statistical analytical approach to predict the secondary structure of proteins from amino acid sequence information

A statistical analytical approach has been used to analyze the secondary structure (SS) of amino acids as a function of the sequence of amino acid residues. We have used 306 non-homologous best-resolved protein structures from the Protein Data Bank for the analysis. A sequence region of 32 amino acids on either side of the residue is considered in order to calculate single amino acid propensities, di-amino acid potentials and tri-amino acid potentials. A weighted sum of predictions obtained using these properties is used to suggest a final prediction method. Our method is as good as the best-known SS prediction methods, is the simplest of all the methods, and uses no homologous sequence/family alignment data, yet gives 72% SS prediction accuracy. Since the method did not use many other factors that may increase the prediction accuracy there is scope to achieve greater accuracy using this approach. Received: 4 May 1998 / Accepted: 17 September 1998 / Published online: 10 December 1998     

Incorporation of FT-IR spectral data in a computer-assisted prediction of globular protein structure

Currently, much effort is being directed to the determination of the three-dimensional structure of proteins. Two classes of research are of interest; spectrometric techniques which include Fourier transform infrared (FT-IR) spectrometry, and non-spectrometric prediction schemes. The spectra obtained using FT-IR spectrometry, are analyzed to determine the percentages of alpha-helices, beta-pleated sheets, and non-structured coils in a protein. Unfortunately, FT-IR, as well as other spectrometric techniques, cannot be used to determine the exact secondary structure of a protein reliably. Non-spectrometric prediction methods yield information on the exact secondary structure, but are not always accurate. Most prediction methods relate the primary amino acid sequence to the secondary structure of a protein, allowing sequential secondary structure information for the protein examined to be obtained. The goal of this research is to incorporate FT-IR with a prediction method, resulting in an improvement in the accuracy of the prediction.     

SPINE X: improving protein secondary structure prediction by multistep learning coupled with prediction of solvent accessible surface area and backbone torsion angles

Accurate prediction of protein secondary structure is essential for accurate sequence alignment, three-dimensional structure modeling, and function prediction. The accuracy of ab initio secondary structure prediction from sequence, however, has only increased from around 77 to 80% over the past decade. Here, we developed a multistep neural-network algorithm by coupling secondary structure prediction with prediction of solvent accessibility and backbone torsion angles in an iterative manner. Our method called SPINE X was applied to a dataset of 2640 proteins (25% sequence identity cutoff) previously built for the first version of SPINE and achieved a 82.0% accuracy based on 10-fold cross validation (Q(3)). Surpassing 81% accuracy by SPINE X is further confirmed by employing an independently built test dataset of 1833 protein chains, a recently built dataset of 1975 proteins and 117 CASP 9 targets (critical assessment of structure prediction techniques) with an accuracy of 81.3%, 82.3% and 81.8%, respectively. The prediction accuracy is further improved to 83.8% for the dataset of 2640 proteins if the DSSP assignment used above is replaced by a more consistent consensus secondary structure assignment method. Comparison to the popular PSIPRED and CASP-winning structure-prediction techniques is made. SPINE X predicts number of helices and sheets correctly for 21.0% of 1833 proteins, compared to 17.6% by PSIPRED. It further shows that SPINE X consistently makes more accurate prediction in helical residues (6%) without over prediction while PSIPRED makes more accurate prediction in coil residues (3-5%) and over predicts them by 7%. SPINE X Server and its training/test datasets are available at http://sparks.informatics.iupui.edu/     

HomologyPlot: Searching for homology to a family of proteins using a database of unique conserved patterns

Summary A new database of conserved amino acid residues is derived from the multiple sequence alignment of over 84 families of protein sequences that have been reported in the literature. This database contains sequences of conserved hydrophobic core patterns which are probably important for structure and function, since they are conserved for most sequences in that family. This database differs from other single-motif or signature databases reported previously, since it contains multiple patterns for each family. The new database is used to align a new sequence with the conserved regions of a family. This is analogous to reports in the literature where multiple sequence alignments are used to improve a sequence alignment. A program called Homology-Plot (suitable for IBM or compatible computers) uses this database to find homology of a new sequence to a family of protein sequences. There are several advantages to using multiple patterns. First, the program correctly identifies a new sequence as a member of a known family. Second, the search of the entire database is rapid and requires less than one minute. This is similar to performing a multiple sequence alignment of a new sequence to all of the known protein family sequences. Third, the alignment of a new sequence to family members is reliable and can reproduce the alignment of conserved regions already described in the literature. The speed and efficiency of this method is enhanced, since there is no need to score for insertions or deletions as is done in the more commonly used sequence alignment methods. In this method only the patterns are aligned. HomologyPlot also provides general information on each family, as well as a listing of patterns in a family.     

Computational tools for protein modeling

Protein modeling is playing a more and more important role in protein and peptide sciences due to improvements in modeling methods, advances in computer technology, and the huge amount of biological data becoming available. Modeling tools can often predict the structure and shed some light on the function and its underlying mechanism. They can also provide insight to design experiments and suggest possible leads for drug design. This review attempts to provide a comprehensive introduction to major computer programs, especially on-line servers, for protein modeling. The review covers the following aspects: (1) protein sequence comparison, including sequence alignment/search, sequence-based protein family classification, domain parsing, and phylogenetic classification; (2) sequence annotation, including annotation/prediction of hydrophobic profiles, transmembrane regions, active sites, signaling sites, and secondary structures; (3) protein structure analysis, including visualization, geometry analysis, structure comparison/classification, dynamics, and electrostatics; (4) three-dimensional structure prediction, including homology modeling, fold recognition using threading, ab initio prediction, and docking. We will address what a user can expect from the computer tools in terms of their strengths and limitations. We will also discuss the major challenges and the future trends in the field. A collection of the links of tools can be found at http://compbio.ornl.gov/structure/resource/.     

Prediction of homology model quality with multivariate regression

A new method has been developed for prediction of homology model quality directly from the sequence alignment, using multivariate regression. Hence, the expected quality of future homology models can be estimated using only information about the primary structure. This method has been applied to protein kinases and can easily be extended to other protein families. Homology model quality for a reference set of homology models was verified by comparison to experimental structures, by calculation of root-mean-square deviations (RMSDs) and comparison of interresidue contact areas. The homology model quality measures were then used as dependent variables in a Partial Least Squares (PLS) regression, using a matrix of alignment score profiles found from the Point Accepted Mutation (PAM) 250 similarity matrix as independent variables. This resulted in a regression model that can be used to predict the accuracy of future homology models from the sequence alignment. Using this method, one can identify the target-template combinations that are most likely to give homology models of sufficient quality. Hence, this method can be used to effectively choose the optimal templates to use for the homology modeling. The method's ability to guide the choice of homology modeling templates was verified by comparison of success rates to those obtained using BLAST scores and target-template sequence identities, respectively. The results indicate that the method presented here performs best in choosing the optimal homology modeling templates. Using this method, the optimal template was chosen in 86% of the cases, as compared to 62% using BLAST scores, and 57% using sequence identities. The method presented here can also be used to identify regions of the protein structure that are difficult to model, as well as alignment errors. Hence, this method is a useful tool for ensuring that the best possible homology model is generated.     

Protein secondary structure prediction with SPARROW

A first step toward predicting the structure of a protein is to determine its secondary structure. The secondary structure information is generally used as starting point to solve protein crystal structures. In the present study, a machine learning approach based on a complete set of two-class scoring functions was used. Such functions discriminate between two specific structural classes or between a single specific class and the rest. The approach uses a hierarchical scheme of scoring functions and a neural network. The parameters are determined by optimizing the recall of learning data. Quality control is performed by predicting separate independent test data. A first set of scoring functions is trained to correlate the secondary structures of residues with profiles of sequence windows of width 15, centered at these residues. The sequence profiles are obtained by multiple sequence alignment with PSI-BLAST. A second set of scoring functions is trained to correlate the secondary structures of the center residues with the secondary structures of all other residues in the sequence windows used in the first step. Finally, a neural network is trained using the results from the second set of scoring functions as input to make a decision on the secondary structure class of the residue in the center of the sequence window. Here, we consider the three-class problem of helix, strand, and other secondary structures. The corresponding prediction scheme "SPARROW" was trained with the ASTRAL40 database, which contains protein domain structures with less than 40% sequence identity. The secondary structures were determined with DSSP. In a loose assignment, the helix class contains all DSSP helix types (α, 3-10, π), the strand class contains β-strand and β-bridge, and the third class contains the other structures. In a tight assignment, the helix and strand classes contain only α-helix and β-strand classes, respectively. A 10-fold cross validation showed less than 0.8% deviation in the fraction of correct structure assignments between true prediction and recall of data used for training. Using sequences of 140,000 residues as a test data set, 80.46% ± 0.35% of secondary structures are predicted correctly in the loose assignment, a prediction performance, which is very close to the best results in the field. Most applications are done with the loose assignment. However, the tight assignment yields 2.25% better prediction performance. With each individual prediction, we also provide a confidence measure providing the probability that the prediction is correct. The SPARROW software can be used and downloaded on the Web page http://agknapp.chemie.fu-berlin.de/sparrow/ .     

A new protein graph model for function prediction

As several structural proteomic projects are producing an increasing number of protein structures with unknown function, methods that can reliably predict protein functions from protein structures are in urgent need. In this paper, we present a method to explore the clustering patterns of amino acids on the 3-dimensional space for protein function prediction. First, amino acid residues on a protein structure are clustered into spatial groups using hierarchical agglomerative clustering, based on the distance between them. Second, the protein structure is represented using a graph, where each node denotes a cluster of amino acids. The nodes are labeled with an evolutionary profile derived from the multiple alignment of homologous sequences. Then, a shortest-path graph kernel is used to calculate similarities between the graphs. Finally, a support vector machine using this graph kernel is used to train classifiers for protein function prediction. We applied the proposed method to two separate problems, namely, prediction of enzymes and prediction of DNA-binding proteins. In both cases, the results showed that the proposed method outperformed other state-of-the-art methods.     

ProteinUnet—An efficient alternative to SPIDER3‐single for sequence‐based prediction of protein secondary structures

Predicting protein function and structure from sequence remains an unsolved problem in bioinformatics. The best performing methods rely heavily on evolutionary information from multiple sequence alignments, which means their accuracy deteriorates for sequences with a few homologs, and given the increasing sequence database sizes requires long computation times. Here, a single‐sequence‐based prediction method is presented, called ProteinUnet, leveraging an U‐Net convolutional network architecture. It is compared to SPIDER3‐Single model, based on long short‐term memory‐bidirectional recurrent neural networks architecture. Both methods achieve similar results for prediction of secondary structures (both three‐ and eight‐state), half‐sphere exposure, and contact number, but ProteinUnet has two times fewer parameters, 17 times shorter inference time, and can be trained 11 times faster. Moreover, ProteinUnet tends to be better for short sequences and residues with a low number of local contacts. Additionally, the method of loss weighting is presented as an effective way of increasing accuracy for rare secondary structures.     

Protein fold recognition

Summary An important, yet seemingly unattainable, goal in structural molecular biology is to be able to predict the native three-dimensional structure of a protein entirely from its amino acid sequence. Prediction methods based on rigorous energy calculations have not yet been successful, and best results have been obtained from homology modelling and statistical secondary structure prediction. Homology modelling is limited to cases where significant sequence similarity is shared between a protein of known structure and the unknown. Secondary structure prediction methods are not only unreliable, but also do not offer any obvious route to the full tertiary structure. Recently, methods have been developed whereby entire protein folds are recognized from sequence, even where little or no sequence similarity is shared between the proteins under consideration. In this paper we review the current methods, including our own, and in particular offer a historical background to their development. In addition, we also discuss the future of these methods and outline the developments under investigation in our laboratory.     

An evaluation of molecular models of the cytochrome P450 Streptomyces griseolus enzymes P450SU1 and P450SU2

Summary P450SU1 and P450SU2 are herbicide-inducible bacterial cytochrome P450 enzymes from Streptomyces griseolus. They have two of the highest sequence identities to camphor hydroxylase (P450cam from Pseudomonas putida), the cytochrome P450 with the first known crystal structure. We have built several models of these two proteins to investigate the variability in the structures that can occur from using different modeling protocols. We looked at variability due to alignment methods, backbone loop conformations and refinement methods. We have constructed two models for each protein using two alignment algorithms, and then an additional model using an identical alignment but different loop conformations for both buried and surface loops. The alignments used to build the models were created using the Needleman-Wunsch method, adapted for multiple sequences, and a manual method that utilized both a dotmatrix search matrix and the Needleman-Wunsch method. After constructing the initial models, several energy minimization methods were used to explore the variability in the final models caused by the choice of minimization techniques. Features of cytochrome P450cam and the cytochrome P450 superfamily, such as the ferredoxin binding site, the heme binding site and the substrate binding site were used to evaluate the validity of the models. Although the final structures were very similar between the models with different alignments, active-site residues were found to be dependent on the conformations of buried loops and early stages of energy minimization. We show which regions of the active site are the most dependent on the particular methods used, and which parts of the structures seem to be independent of the methods.     

MSAID: multiple sequence alignment based on a measure of information discrepancy

We propose an algorithm of global multiple sequence alignment that is based on a measure of what we call information discrepancy. The algorithm follows a progressive alignment iteration strategy that makes use of what we call a function of degree of disagreement (FDOD). MSAID begins with distance calculation of pairwise sequences, based on FDOD as a numerical scoring measure. In the next step, the resulting distance matrix is used to construct a guide tree via the neighbor-joining method. The tree is then used to produce a multiple alignment. Current alignment is next used to produce a new matrix and a new tree (with FDOD scoring measure again). This iterative process continues until convergence criteria (or a stopping rule) are satisfied. MSAID was tested and compared with other prior methods by using reference alignments from BAliBASE 2.01. For the alignments with no large N/C-terminal extensions or internal insertions MSAID received the top overall average in the tests. Moreover, the results of testing indicate that MSAID performs as well as other alignment methods with an occasional tendency to perform better than these prior techniques. We, therefore, believe that MSAID is a solid and reliable method of choice, which is often (if not always) superior to other global alignment techniques.     

Inter-domain linker prediction using amino acid compositional index

Protein chains are generally long and consist of multiple domains. Domains are distinct structural units of a protein that can evolve and function independently. The accurate and reliable prediction of protein domain linkers and boundaries is often considered to be the initial step of protein tertiary structure and function predictions. In this paper, we introduce CISA as a method for predicting inter-domain linker regions solely from the amino acid sequence information. The method first computes the amino acid compositional index from the protein sequence dataset of domain-linker segments and the amino acid composition. A preference profile is then generated by calculating the average compositional index values along the amino acid sequence using a sliding window. Finally, the protein sequence is segmented into intervals and a simulated annealing algorithm is employed to enhance the prediction by finding the optimal threshold value for each segment that separates domains from inter-domain linkers. The method was tested on two standard protein datasets and showed considerable improvement over the state-of-the-art domain linker prediction methods.     

In silico Protein-Protein Interaction prediction with sequence alignment and classifier stacking

Protein-Protein Interaction (PPI) prediction is a well known problem in Bioinformatics, for which a large number of techniques have been proposed in the past. However, prediction results have not been sufficiently satisfactory for guiding biologists in web-lab experiments. One reason is that not all useful information, such as pairwise protein interaction information based on sequence alignment, has been integrated together in PPI prediction. Alignment is a basic concept to measure sequence similarity in Proteomics that has been used in a number of applications ranging from protein recognition to protein subcellular localization. In this article, we propose a novel integrated approach to predicting PPI based on sequence alignment by jointly using a k-Nearest Neighbor classifier (SA-kNN) and a Support Vector Machine (SVM). SVM is a machine learning technique used in a wide range of Bioinformatics applications, thanks to the ability to alleviate the overfitting problems. We demonstrate that in our approach the two methods, SA-kNN and SVM, are complementary, which are combined in an ensemble to overcome their respective limitations. While the SVM is trained on Amino Acid (AA) compositions and protein signatures mined from literature, the SA-kNN makes use of the similarity of two protein pairs through alignment. Experimentally, our technique leads to a significant gain in accuracy, precision and sensitivity measures at ~5%, 16% and 10% respectively.     

An alignment-free method to find similarity among protein sequences via the general form of Chou’s pseudo amino acid composition

In this paper, we propose a method to create the 60-dimensional feature vector for protein sequences via the general form of pseudo amino acid composition. The construction of the feature vector is based on the contents of amino acids, total distance of each amino acid from the first amino acid in the protein sequence and the distribution of 20 amino acids. The obtained cosine distance metric (also called the similarity matrix) is used to construct the phylogenetic tree by the neighbour joining method. In order to show the applicability of our approach, we tested it on three proteins: 1) ND5 protein sequences from nine species, 2) ND6 protein sequences from eight species, and 3) 50 coronavirus spike proteins. The results are in agreement with known history and the output from the multiple sequence alignment program ClustalW, which is widely used. We have also compared our phylogenetic results with six other recently proposed alignment-free methods. These comparisons show that our proposed method gives a more consistent biological relationship than the others. In addition, the time complexity is linear and space required is less as compared with other alignment-free methods that use graphical representation. It should be noted that the multiple sequence alignment method has exponential time complexity.     

Comparative analysis of threshold and tessellation methods for determining protein contacts

The 3D structure of a protein is the main physical support of a protein's biological function; 3D protein folds are primarily maintained through interactions between amino acids. Inter-residue contacts are essential for the stability of protein folds. Therefore, many methodologies in the fields of structure analysis, structure prediction, and structure-function relationships are based on residue contacts. The present study provides a comparative analysis of two approaches for determining contacts: the classical distance-threshold method and an application of Laguerre, or weighted Voronoi tessellation. First, we examined mean contact distributions and their dependence on residue volumes, accessibility and hydrophobicity. In general, the different methods gave concordant results, although the method based on Cα distances showed significant discrepancies with the all-atom tessellation method. We also analyzed preferential contacts between all amino acid species and studied the influence of protein chain length, the proximity of the residues along the sequence, and the secondary structure environment. Interestingly, the discrepancies between methods were occasionally large enough to substantially change the relative preferences of some contacts. Finally, a case study on disulfide bridges demonstrated the importance of the structural environment in determining contacts from tessellation. In conclusion, the tessellation method is more accurate because of its fine adaptation to local protein topology, with far-reaching implications for most contact-based prediction methods of protein folding.     

PreSSAPro: a software for the prediction of secondary structure by amino acid properties

PreSSAPro is a software, available to the scientific community as a free web service designed to provide predictions of secondary structures starting from the amino acid sequence of a given protein. Predictions are based on our recently published work on the amino acid propensities for secondary structures in either large but not homogeneous protein data sets, as well as in smaller but homogeneous data sets corresponding to protein structural classes, i.e. all-alpha, all-beta, or alpha–beta proteins. Predictions result improved by the use of propensities evaluated for the right protein class. PreSSAPro predicts the secondary structure according to the right protein class, if known, or gives a multiple prediction with reference to the different structural classes. The comparison of these predictions represents a novel tool to evaluate what sequence regions can assume different secondary structures depending on the structural class assignment, in the perspective of identifying proteins able to fold in different conformations. The service is available at the URL http://bioinformatica.isa.cnr.it/PRESSAPRO/.     

蛋白质折叠类型的分类建模与识别

蛋白质的氨基酸序列如何决定空间结构是当今生命科学研究中的核心问题之一. 折叠类型反映了蛋白质核心结构的拓扑模式, 折叠识别是蛋白质序列-结构研究的重要内容. 我们以占Astral 1.65序列数据库中α, β和α/β三类蛋白质总量41.8%的36个无法独立建模的折叠类型为研究对象, 选取其中序列一致性小于25%的样本作为训练集, 以均方根偏差(RMSD)为指标分别进行系统聚类, 生成若干折叠子类, 并对各子类建立基于多结构比对算法(MUSTANG)结构比对的概形隐马尔科夫模型(profile-HMM). 将Astral 1.65中序列一致性小于95%的9505个样本作为检验集, 36个折叠类型的平均识别敏感性为90%, 特异性为99%, 马修斯相关系数(MCC)为0.95. 结果表明: 对于成员较多, 无法建立统一模型的折叠类型, 基于RMSD的系统分类建模均可实现较高准确率的识别, 为蛋白质折叠识别拓展了新的方法和思路, 为进一步研究奠定了基础.