Many online chat applications live in a grey area between the legitimate web and the dark net. The Telegram network in particular can aid criminal activities. Telegram hosts “chats” which consist of varied conversations and advertisements. These chats take place among automated “bots” and human users. Classifying legitimate activity from illegitimate activity can aid law enforcement in finding criminals. Social network analysis of Telegram chats presents a difficult problem. Users can change their username or create new accounts. Users involved in criminal activity often do this to obscure their identity. This makes establishing the unique identity behind a given username challenging. Thus we explored classifying users from their language usage in their chat messages.The volume and velocity of Telegram chat data place it well within the domain of big data. Machine learning and natural language processing (NLP) tools are necessary to classify this chat data. We developed NLP tools for classifying users and the chat group to which their messages belong. We found that legitimate and illegitimate chat groups could be classified with high accuracy. We also were able to classify bots, humans, and advertisements within conversations.

BACKGROUND: Drug resistance is a critical problem limiting effective antiviral therapy for HIV/AIDS. Computational techniques for predicting drug resistance profiles from genomic data can accelerate the appropriate choice of therapy. These techniques can also be used to identify protease mutants for experimental studies of resistance and thereby assist in the development of next-generation therapies. Few studies, however, have assessed the evolution of resistance from genotype-phenotype data. RESULTS: The machine learning produced highly accurate and robust classification of resistance to HIV protease inhibitors. Genotype data were mapped to the enzyme structure and encoded using Delaunay triangulation. Estimates of evolutionary relationships, based on this encoding, and using Minimum Spanning Trees, showed clusters of mutations that closely resemble the wild type. These clusters appear to evolve uniquely to more resistant phenotypes. CONCLUSIONS: Using the triangulation metric and spanning trees results in paths that are consistent with evolutionary theory. The majority of the paths show bifurcation, namely they switch once from non-resistant to resistant or from resistant to non-resistant. Paths that lose resistance almost uniformly have far lower levels of resistance than those which either gain resistance or are stable. This strongly suggests that selection for stability in the face of a rapid rate of mutation is as important as selection for resistance in retroviral systems.

Network based attacks are the major threat to security on the Internet. The volume of traffic and the high variability of the attacks place threat detection squarely in the domain of big data. Conventional approaches are mostly based on signatures. While these are relatively inexpensive computationally, they are inflexible and insensitive to small variations in the attack vector. Therefore we explored the use of machine learning techniques on real flow data. We found that benign traffic could be identified with high accuracy.

BACKGROUND: Drug resistance in HIV is the major problem limiting effective antiviral therapy. Computational techniques for predicting drug resistance profiles from genomic data can accelerate the appropriate choice of therapy. These techniques can also be used to select protease mutants for experimental studies of resistance and thereby assist in the development of next-generation therapies. RESULTS: The machine learning produced highly accurate and robust classification of HIV protease resistance. Genotype data were mapped to the enzyme structure and encoded using Delaunay triangulation. Generative machine learning models trained on one inhibitor could classify resistance from other inhibitors with varying levels of accuracy. Generally, the accuracy was best when the inhibitors were chemically similar. CONCLUSIONS: Restricted Boltzmann Machines are an effective machine learning tool for classification of genomic and structural data. They can also be used to compare resistance profiles of different protease inhibitors.

BACKGROUND: HIV/AIDS is a serious threat to public health. The emergence of drug resistance mutations diminishes the effectiveness of drug therapy for HIV/AIDS. Developing a computational prediction of drug resistance phenotype will enable efficient and timely selection of the best treatment regimens. RESULTS: A unified encoding of protein sequence and structure was used as the feature vector for predicting phenotypic resistance from genotype data. Two machine learning algorithms, Random Forest and K-nearest neighbor, were used. The prediction accuracies were examined by five-fold cross-validation on the genotype-phenotype datasets. A supervised machine learning approach for automatic prediction of drug resistance was developed to handle genotype-phenotype datasets of HIV protease (PR) and reverse transcriptase (RT). It predicts the drug resistance phenotype and its relative severity from a query sequence. The accuracy of the classification was higher than 0.973 for eight PR inhibitors and 0.986 for ten RT inhibitors, respectively. The overall cross-validated regression R(2)-values for the severity of drug resistance were 0.772-0.953 for 8 PR inhibitors and 0.773-0.995 for 10 RT inhibitors. CONCLUSIONS: Machine learning using a unified encoding of sequence and protein structure as a feature vector provides an accurate prediction of drug resistance from genotype data. A practical webserver for clinicians has been implemented.

BACKGROUND: Drug resistance is one of the most important causes for failure of anti-AIDS treatment. During therapy, multiple mutations accumulate in the HIV genome, eventually rendering the drugs ineffective in blocking replication of the mutant virus. The huge number of possible mutants precludes experimental analysis to explore the molecular mechanisms of resistance and develop improved antiviral drugs. RESULTS: In order to solve this problem, we have developed a new algorithm to reveal the most representative mutants from the whole drug resistant mutant database based on our newly proposed unified protein sequence and 3D structure encoding method. Mean shift clustering and multiple regression analysis were applied on genotype-resistance data for mutants of HIV protease and reverse transcriptase. This approach successfully chooses less than 100 mutants with the highest resistance to each drug out of about 10K in the whole database. When considering high level resistance to multiple drugs, the numbers reduce to one or two representative mutants. CONCLUSION: This approach for predicting the most representative mutants for each drug has major importance for experimental verification since the results provide a small number of representative sequences, which will be amenable for in vitro testing and characterization of the expressed mutant proteins.

Effective Big Data applications dynamically handle the retrieval of decisioned results based on stored large datasets efficiently. One effective method of requesting decisioned results, or querying, large datasets is the use of SQL and database management systems such as MySQL. But a problem with using relational databases to store huge datasets is the decisioned result retrieval time, which is often slow largely due to poorly written queries / decision requests. This work presents a model to re-architect Big Data applications in order to efficiently present decisioned results: lowering the volume of data being handled by the application itself, and significantly decreasing response wait times while allowing the flexibility and permanence of a standard relational SQL database, supplying optimal user satisfaction in today's Data Analytics world. In this paper we review a Big Data case study in the telecommunications field and use it to experimentally demonstrate the effectiveness of our approach.

The determination of whether a molecule can bind or ”dock” a protein to a certain site depends on the orientation of the molecules, the charge of the atoms that comprise the molecules, and the electrical potential of the proposed area. It is said that a fundamental problem with molecular docking is that the orientation space itself is very large and grows in a combinatorial manner with the number of degrees of freedom of the interacting molecules. We tried to cleverly solve this problem by using shape definitions and fuzzy logic to help reduce the search size of possible docking locations and predict which locations and orientations are the most likely docking locations.

Effective Big Data applications dynamically handle the retrieval of decisioned results based on stored large datasets efficiently. One effective method of requesting decisioned results, or querying, large datasets is the use of SQL and database management systems such as MySQL. But a problem with using relational databases to store huge datasets is the decisioned result retrieval time, which is often slow largely due to poorly written queries/decision requests. This work presents a model to re-architect Big Data applications in order to efficiently present decisioned results: lowering the volume of data being handled by the application itself, and significantly decreasing response wait times while allowing the flexibility and permanence of a standard relational SQL database, supplying optimal user satisfaction in today's Data Analytics world. We experimentally demonstrate the effectiveness of our approach.

HIV rapidly evolves drug resistance in response to antiviral drugs used in AIDS therapy. Estimating the specific resistance of a given strain of HIV to individual drugs from sequence data has important benefits for both the therapy of individual patients and the development of novel drugs. We have developed an accurate classification method based on the sparse representation theory, and demonstrate that this method is highly effective with HIV-1 protease. The protease structure is represented using our newly proposed encoding method based on Delaunay triangulation, and combined with the mutated amino acid sequences of known drug-resistant strains to train a machine-learning algorithm both for classification and regression of drug-resistant mutations. An overall cross-validated classification accuracy of 97% is obtained when trained on a publically available data base of approximately 1.5 × 104 known sequences (Stanford HIV database: http://hivdb.stanford.edu/cgi-bin/GenoPhenoDS.cgi). Resistance to four FDA approved drugs is computed and comparisons with other algorithms demonstrate that our method shows significant improvements in classification accuracy.

Automatic prediction of protein three dimensional structures from its amino acid sequence has become one of the most important and researched fields in bioinformatics We attempt to solve this problem using machine learning technique and information from both sequence and structure of the protein. Information like amino acid substitution matrix, polarity, secondary structure information and relative distance between alpha carbon atoms etc is collected through spatial traversing of the 3D structure to form training vectors. This guarantees that the properties of alpha carbon atoms that are close together in 3D space and thus interacting are used in vector formation. The goal is to generate a machine that understands structures from PDB and when given a new model, predicts whether it belongs to the same class as the PDB structures (correct or incorrect protein models). Improved fuzzy decision tree algorithm is used to build the machine, for the rules generated and high prediction accuracy it is favored over other machine learning techniques. Different subsets of PDB are considered for evaluating the prediction potential of the machine learning methods. With the use of machine learning technique, fuzzy decision tree, we obtained a training accuracy around 90%. There is significant improvement compared to previous encoding technique in prediction accuracy and execution time. This outcome motivates to continue to explore effective machine learning algorithms for accurate protein model quality assessment.

Computational systems biology is largely driven by mathematical modeling and simulation of biochemical networks, via continuous deterministic methods or discrete event stochastic methods. Although the deterministic methods are efficient in predicting the macroscopic behavior of a biochemical system, they are severely limited by their inability to represent the stochastic effects of random molecular fluctuations at lower concentration. In this work, we have presented a novel method for simulating biochemical networks based on a deterministic solution with a modification that permits the incorporation of stochastic effects. To demonstrate the feasibility of our approach, we have tested our method on three previously reported biochemical networks. The results, while staying true to their deterministic form, also reflect the stochastic effects of random fluctuations that are dominant as the system transitions into a lower concentration. This ability to adapt to a concentration gradient makes this method particularly attractive for systems biology-based applications.

MicroRNAs play an important role in post-transcriptional gene regulation. Experimental approaches to identify microRNAs are expensive and time-consuming. Computational approaches have proven to be useful for identifying microRNA candidates. Most approaches rely on features extracted from miroRNA precursors (pre-microRNA) and their secondary structure. Selecting the appropriate set of features plays a critical role in improving the prediction accuracy of pre-microRNA candidates. This work aims to investigate the triplet elements encoding scheme and to identify essential features needed for the correct classification of pre-microRNAs. To achieve these goals, an extension of the triplet elements encoding scheme is introduced. Features extracted using the extended scheme were combined with global features introduced in the literature, and fuzzy decision tree (FDT) is used as a classification and a feature selection tool. Unlike previous machine-learning-based approaches, FDT produces a human comprehensible classification model. The interpretability of the classification model provides a means to identify the essential features needed to recognize microRNA candidates and offers a better understanding of this problem. Our results indicate that the triplet elements scheme is not superior to any of its proposed extensions. Further analysis revealed that including the features extracted using triplet elements scheme does not add any value to this classification problem but rather introduces some noisy features, and comparable classification results can be achieved by using only the six global features identified by FDT.

Blood testosterone levels oscillate periodically in humans. The in vivo dynamics of this biochemical system cannot be simulated in silico using a continuous deterministic solution of a previously reported mathematical model. The use of the stochastic simulation algorithm (SSA), however, has been reported to generate sustained oscillations that are qualitatively and quantitatively consistent with the experimental observations. Although the SSA is capable of accurately simulating a biochemical network, it is extremely inefficient from a computational standpoint. In this work, we have attempted to simulate the above mentioned model using a deterministic-stochastic crossover method, for three separate sets of parameters. Each time, not only did the results show the existence of sustained oscillations but also that the computational time was at least four times lower than the corresponding SSA solution. The crossover method can hence be proposed as a viable alternative to the SSA for simulating biochemical systems that are commonly encountered in systems biology applications.

Recent work using supervised learning for protein structure prediction has achieved state-of-the-art classification performance. However, such methods are based only on labeled data, while in practice the labeled data is so few and expensive to obtain and unlabeled data is far more plentiful. An effective way to enhance the performance of the learned hypothesis by using the labeled and unlabeled data together is known as semi-supervised learning. Although there are lots of semi-supervised learning methods, those approaches could not always achieve the acceptable results for bioinformatics application, especially when there is only very few labeled instances. Therefore, in this paper, we present a novel, more effective method tri-cluster-tri-scheme-training (TCTS) which firstly uses tri-cluster to label some high confidence unlabeled instances and then refines the classifiers by utilizing both of the label data and unlabeled data in the Tri-Scheme-training by different schemes. The encouraging experimental results indicate that TCTS algorithm opens a new way to solve the complex classification problem when very few labeled datasets are available.

Computational systems biology is largely driven by mathematical modeling and simulation of biochemical networks, via continuous deterministic methods or discrete event stochastic methods. Although, the deterministic methods are efficient in predicting the macroscopic behavior of a system, they are severely limited by their inability to represent the stochastic effects of random molecular fluctuations at lower concentration. In this work, we have presented a novel method for simulating biochemical networks based on a deterministic solution with a modification that permits the incorporation of stochastic effects. To demonstrate the feasibility of our approach, we have tested our method on two previously reported biochemical networks. The results, while staying true to their deterministic form, also reflect the stochastic effects of random fluctuations that are dominant as the system transitions into a lower concentration. This ability to adapt to a concentration gradient makes this method particularly attractive for systems biology based applications.

Decision-tree algorithms are one of the most popular applications in machine learning. The ID3 algorithm is an efficient method for building decision trees that form the basis for many decision tree programs. Fuzzy ID3 is an extension of the existing ID3 algorithm; it integrates fuzzy set theory and ID3 to overcome the effects of spurious precision in the data, to treat uncertainties in the data and to reduce the decision tree sensitivity to small changes in attribute values. In this paper, we introduce a modified version of fuzzy ID3 algorithm that integrates information gain and classification ambiguity to select the test attribute. The modified algorithm achieves better accuracy than the original Fuzzy ID3 as well as crisp programs such C4.5 on a wide range of datasets. We also introduce a new machine learning software tool based on fuzzy decision trees.

SUMMARY: The accumulation of high-throughput genomic, proteomic and metabolical data allows for increasingly accurate modeling and reconstruction of metabolic networks. Alignment of the reconstructed networks can help to catch model inconsistencies and infer missing elements. In this note, we present the web service tool MetNetAligner which aligns metabolic networks, taking in account the similarity of network topology and the enzymes' functions. It can be used for predicting unknown pathways, comparing and finding conserved patterns and resolving ambiguous identification of enzymes. The tool supports several alignment options including allowing or forbidding enzyme deletion and insertion. It is based on a novel scoring scheme which measures enzyme-to-enzyme functional similarity and a fast algorithm which efficiently finds optimal mappings from a directed graph with restricted cyclic structure to an arbitrary directed graph. AVAILABILITY: MetNetAligner is available as web-server at: http://alla.cs.gsu.edu:8080/MinePW/pages/gmapping/GMMain.html. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Network alignments are extensively used for comparing, exploring, and predicting biological networks. Existing alignment tools are mostly based on isomorphic and homeomorphic embedding and require solving a problem that is NP-complete even when searching a match for a tree in acyclic networks. On the other hand, if the mapping of different nodes from the query network (pattern) into the same node from the text network is allowed, then trees can be optimally mapped into arbitrary networks in polynomial time.In this paper we present the first polynomial-time algorithm for finding the best matching pair consisting of a subtree in a given tree pattern and a subgraph in a given text (represented by an arbitrary network) when both insertions and deletions of degree-2 vertices are allowed on any path. Our dynamic programming algorithm is an order of magnitude faster than the previous network alignment algorithm when deletions are forbidden. The algorithm has been also generalized to pattern networks with cycles: with a modest increase in runtime it can handle patterns with the limited vertex feedback set.We have applied our algorithm to matching metabolic pathways of four organisms (E. coli, S. cerevisiae, B. subtilis and T. thermophilus species) and found a reasonably large set of statistically significant alignments. We show advantages of allowing pattern vertex deletions and give an example validating biological relevance of the pathway alignment.

Protein sequence motifs are gathering more and more attention in the sequence analysis area. These recurring regions have the potential to determine protein 's conformation, function and activities. In our previous work, we tried to obtain protein sequence motifs which are universally conserved across protein family boundaries. Therefore, unlike most popular motif discovering algorithms, our input dataset is extremely large. In order to deal with large input datasets, we provided two granular computing models (FIK and FGK model) to efficiently generate protein motifs information and Super GSVM-FE model to do the feature elimination for improving the quality of motif information. In this article, we tried to further improve our SVM feature elimination model to achieve three goals: Reduce time execution by half, further improve motif information quality and add the ability of adjusting the number of filtered segments. Compared with the latest results, our new approach shows great improvements.

Network mapping is a convenient tool for comparing and exploring biological networks; it can be used for predicting unknown pathways, fast and meaningful searching of databases, and potentially establishing evolutionary relations. Unfortunately, existing tools for mapping paths into general networks (PathBlast) or trees into tree networks allowing gaps (MetaPathwayHunter) cannot handle large query pathways or complex networks. In this paper we consider homomorphisms, i.e., mappings allowing to map different enzymes from the query pathway into the same enzyme from the networks. Homomorphisms are more general than homeomorphism (allowing gaps) and easier to handle algorithmically. Our dynamic programming algorithm efficiently finds the minimum cost homomorphism from a multisource tree to directed acyclic graphs as well as general networks. We have performed pairwise mapping of all pathways for four organisms (E. coli, S. cerevisiae, B. subtilis and T. thermophilus species) and found a reasonably large set of statistically significant pathway similarities. Further analysis of our mappings identifies conserved pathways across examined species and indicates potential pathway holes in existing pathway descriptions.

Protein sequence motifs are gathering more and more attention in the sequence analysis area. These recurring regions have the potential to determine protein's conformation, function and activities. In our previous work, we tried to obtain protein sequence motifs which are universally conserved across protein family boundaries. Therefore, unlike most popular motif discovering algorithms, our input dataset is extremely large. In order to deal with large input datasets, we provided two granular computing models (FIK and FGK model) to efficiently generate protein motifs information. In this article, we develop a new method which combines the concept of granular computing and the power of ranking SVM to further extract protein sequence motif information. There are two reasons to eliminate redundant data: First, the information we try to generate is about sequence motifs, but the original input data are derived from whole protein sequences by a sliding window technique; second, during fuzzy c-means clustering, it has the ability to assign one segment to more than one information granule. However, not all data segments have a direct relation to the granule they assigned. The quality of motif information increases dramatically in all three evaluation measures by applying this new feature elimination model. Compared with traditional methods which shrink cluster size to obtain a more compact one, our approach shows improved results.

With the efforts to understand protein structure, many computational approaches have been made recently. Among them, the support vector machine (SVM) methods have been recently applied and showed successful performance compared with other machine learning schemes. However, despite the high performance, the SVM approaches suffer from the problem of understandability since it is a black-box model. To overcome this limitation, this study attempted to combine the SVM with the association rule based classifier which can present the meaningful explanation about the prediction. To perform this task, a new association rule based classifier (PCPAR) was devised based on the existing classifier, CPAR, to handle the sequential data. PCPAR creates the patterns by merging the generated rules and then classifies the sequential data based on the pattern match. The experimental result presents the following: with sequential data, the PCPAR scheme shows better performance with respect to the accuracy and the number of generated patterns than CPAR method whether applied alone or combined with SVM. The combined scheme of SVMPCPAR generates more compact patterns than the combined scheme of SVM with decision tree, SVM DT, with similar performance. These patterns are easily understandable and biologically meaningful

We present a novel design for managing competing user interests and privacy protection cooperation in a collaborative virtual environment for molecular modeling. Based on the user feedback, this design includes four levels of access control for collaborative sessions and provides dynamic action priority specification for manipulations on shared molecular models. Furthermore, we implement a messaging system that includes a text chatting tool and system broadcasting functions. Our design supports flexible user group formation and interaction. We have implemented our design on a set of Microsoft Windows PCs connected to a Linux molecule simulation server. A GUI interface developed in Python is also illustrated

Protein sequence motifs information is very important to the analysis of biologically significant regions. The conserved regions have the potential to determine the conformation, function and activities of the proteins. The main purpose of this paper is trying to obtain protein sequence motifs which are universally conserved and across protein family boundaries. Therefore, unlike most popular motif discovering algorithms, our input dataset is extremely large. As a result, an efficient technique is demanded. In this article, short recurring segments of proteins are explored by utilizing a novel granular computing strategy. First, fuzzy C-means clustering algorithm (FCM) is used to separate the whole dataset into several smaller informational granules and then succeeded by improved K-means clustering algorithm on each granule to obtain the final results. The structural similarity of the clusters discovered by our approach is studied to analyze how the recurring patterns correlate with its structure. Also, some biochemical references are included in our evaluation. To the best of our knowledge, it is the first time that the granular computing concept as well as the DBI measure for evaluation is introduced to this dataset. Compare with the latest research results, our method requires only twenty percent of the execution time and obtains even higher quality information of protein sequence motifs. The efficient and satisfactory results in our experiment suggests that our granular computing model which combined FCM and improved K-means may have a high chance to be applied in some other bioinformatics research fields and yield stunning results

Support vector machine (SVM) classification often heavily relies on selected kernel functions. This paper proposes a fuzzy fusion model to combine multi-SVMs to improve the performance of SVM classification. In order to better handle uncertainties in real classification applications, we apply type-2 fuzzy sets to create the fusion model. The model takes the classification results from multi-SVMs and generates the combined decision. Our experiments show the proposed model outperforms individual SVMs in most cases and also has better performance than type-1 based fusion model in general.

RNA plays a critical role in mediating every step of cellular information transfer from genes to functional proteins. Pseudoknots are widely occurring structural motifs found in all types of RNA and are also functionally important. Therefore predicting their structures is an important problem. In this paper, we present a new RNA pseudoknot prediction method based on term rewriting rather than on dynamic programming, comparative sequence analysis, or context-free grammars. The method we describe is implemented using the Mfold RNA/DNA folding package and the term rewriting language Maude. Our method was tested on 211 pseudoknots in PseudoBase and achieves an average accuracy of 74.085% compared to the experimentally determined structure. In fact, most pseudoknots discovered by our method achieve an accuracy of above 90%. These results indicate that term rewriting has a broad potential in RNA applications from prediction of pseudoknots to higher level RNA structures involving complex RNA tertiary interactions.

Combining multiple classifiers is a natural way to discover useful information and improve the performance of individual classifiers. In this paper, we propose one approach to combine multiple SVMs and improve the generalization ability of SVM classifiers. One fuzzy system is constructed based on SVM accuracies and distances of data examples to SVM hyperplanes. The output fuzzy membership functions of the fuzzy system are tuned by a genetic algorithm (GA). The established model is applied on an ovarian cancer dataset and the experiment shows the proposed genetic fuzzy model performs more stable and more reliable than individual SVMs.

Molecular modeling is an important research area, helping scientists develop new drugs against diseases such as AIDS and cancer. Prior studies have demonstrated that immersive virtual environments have unique advantages over desktop systems in visualizing molecular models. However, exploration and interaction in existing molecular modeling virtual environments is often limited to a single user, lacking strong support for collaboration. In addition, scientists are often reluctant to adopt these systems because of their lack of availability and high cost. We propose an affordable immersive system that allows biologists and chemists to manipulate molecular models via natural gestures, receive and visualize real-time feedback from a molecular dynamics simulator, allow the sharing of customized views, and provide support for both local and remote collaborative research.

A new prediction scheme of transmembrane segments (TM) was developed based on the support vector machine (SVM). To apply this SVM for prediction more efficiently, three optimization processes were performed: encoding scheme, sliding window size and parameter optimization. From the encoding scheme optimization, position-specific scoring matrix (PSSM) encoding scheme is proved to be the most informative one and the prediction accuracy (Q/sub 2/) with this scheme attained up to 92%. Based on the performance comparison with previous studies, this PSSM encoding scheme demonstrates the highest prediction accuracy among the common prediction methods, and the accuracy improvement is more than 13%. To verify this scheme, the blind test was done with E.coli SecE and E.coli SecY transmembrane proteins, and the result shows a decent match with the SwissProt database information and the TopPred results. However, another blind test result with five SecA proteins leaves room for discussion since it shows about 8-9 residues long TM segments for all five proteins.

Protein secondary structure prediction is one of the most important problems in bioinformatics research. When the traditional tertiary classifier is used in our neural network, 72% accuracy is reached. Since the neural network might not work very well in three-class classification for certain domains, the three-class problem is reduced to six binary class problems for the first time to carry out protein secondary structure prediction. With the combination of six binary classifiers, we experiment and test several tertiary classifiers. Additionally, three new tertiary classifiers are proposed in this study: MAX/spl I.bar/HEC, ONE/spl I.bar/TO/spl I.bar/ONE/spl I.bar/MAX and ONE/spl I.bar/TO/spl I.bar/ONE/spl I.bar/VOTE. ONE/spl I.bar/TO/spl I.bar/ONE/spl I.bar/VOTE outperforms the six other experimental tertiary classifiers in this study. ONE/spl I.bar/TO/spl I.bar/ONE/spl I.bar/VOTE tertiary classifier with PSSM encoding scheme obtains 74.02% test accuracy on RS126 dataset. To the best of our knowledge, this is the best result for RS126 dataset with the cross-validation method for neural network. The improvement of prediction accuracy indicates that decomposition of the multiclass problem into several binary class problems may be applied to other areas of computational biology in order to increase generalization power of neural networks.