Predicting protein structural class by functional domain composition

Identification of protein functions in mouse with a label space partition method

Protein is very important for almost all living creatures because it participates in most complicated and essential biological processes. Determining the functions of given proteins is one of the most essential problems in protein science. Such determination can be conducted through traditional experiments. However, the experimental methods are always time-consuming and of high costs. In recent years, computational methods give useful aids for identification of protein functions. This study presented a new multi-label classifier for identifying functions of mouse proteins. Due to the number of functional types, which were termed as labels in the classification procedure, a label space partition method was employed to divide labels into some partitions. On each partition, a multi-label classifier was constructed. The classifiers based on all partitions were integrated in the proposed classifier. The cross-validation results proved that the proposed classifier was of good performance. Classifiers with label partition were superior to those without label partition or with random label partition.

Using Recursive Feature Selection with Random Forest to Improve Protein Structural Class Prediction for Low-Similarity Sequences

Many combinations of protein features are used to improve protein structural class prediction, but the information redundancy is often ignored. In order to select the important features with strong classification ability, we proposed a recursive feature selection with random forest to improve protein structural class prediction. We evaluated the proposed method with four experiments and compared it with the available competing prediction methods. The results indicate that the proposed feature selection method effectively improves the efficiency of protein structural class prediction. Only less than 5% features are used, but the prediction accuracy is improved by 4.6-13.3%. We further compared different protein features and found that the predicted secondary structural features achieve the best performance. This understanding can be used to design more powerful prediction methods for the protein structural class.

Recent Advances in the Prediction of Protein Structural Classes: Feature Descriptors and Machine Learning Algorithms

In the postgenomic age, rapid growth in the number of sequence-known proteins has been accompanied by much slower growth in the number of structure-known proteins (as a result of experimental limitations), and a widening gap between the two is evident. Because protein function is linked to protein structure, successful prediction of protein structure is of significant importance in protein function identification. Foreknowledge of protein structural class can help improve protein structure prediction with significant medical and pharmaceutical implications. Thus, a fast, suitable, reliable, and reasonable computational method for protein structural class prediction has become pivotal in bioinformatics. Here, we review recent efforts in protein structural class prediction from protein sequence, with particular attention paid to new feature descriptors, which extract information from protein sequence, and the use of machine learning algorithms in both feature selection and the construction of new classification models. These new feature descriptors include amino acid composition, sequence order, physicochemical properties, multiprofile Bayes, and secondary structure-based features. Machine learning methods, such as artificial neural networks (ANNs), support vector machine (SVM), K-nearest neighbor (KNN), random forest, deep learning, and examples of their application are discussed in detail. We also present our view on possible future directions, challenges, and opportunities for the applications of machine learning algorithms for prediction of protein structural classes.

A two-stage approach towards protein secondary structure classification

Protein secondary structure (PSS) describes the local folded structures which get formed inside a polypeptide due to interactions among atoms of the backbone. Generally, globular proteins are divided into four classes, namely all-α, all-β, α + β, and α/β. As nearly 90% of proteins fall into the said four classes, these are mostly considered for the purpose of computational classification of proteins. Classification of PSS is important for different biological functions that include protein fold recognition, tertiary structure prediction, prediction of DNA-binding sites, and reduction of the conformation search space among others. In this paper, we have proposed a machine learning-based model for secondary structure classification of proteins into four classes: all-α, all-β, α + β, and α/β. In doing so, we have considered both sequence-based and structure-based features. At first, mutual information (MI), a filter-based feature selection method, is used to remove the redundant features, and then these selected features are used to train three different classifiers-random forest, K-nearest neighbor (KNN), and multi-layer perceptron (MLP). After that, some standard classifier combination approaches are applied to integrate the decision made by the said classifiers and it has been found that weighted product rule performs the best among all. The overall accuracies obtained using the proposed model on the four standard datasets, namely 640, 1189, 25pdb, and fc699 are 86.89%, 92.93%, 91.38%, and 94.87% respectively. The proposed model outperforms some state-of-the-art methods considered here for comparison. Significantly high classification accuracy produced by our proposed model on four datasets is attributed to the development of a comprehensive feature set (by eliminating redundant features through feature selection technique) which is then passed through an ensemble consists of three different classifiers. Assigning different weights to the outcome of different classifiers thus proved to be useful in designing the model for predicting the secondary structure of proteins based on its sequence-based and structure-based features. Graphical abstract.

A Gram-Negative Bacterial Secreted Protein Types Prediction Method Based on PSI-BLAST Profile

Prediction of secreted protein types based solely on sequence data remains to be a challenging problem. In this study, we extract the long-range correlation information and linear correlation information from position-specific score matrix (PSSM). A total of 6800 features are extracted at 17 different gaps; then, 309 features are selected by a filter feature selection method based on the training set. To verify the performance of our method, jackknife and independent dataset tests are performed on the test set and the reported overall accuracies are 93.60% and 100%, respectively. Comparison of our results with the existing method shows that our method provides the favorable performance for secreted protein type prediction.

Probing an optimal class distribution for enhancing prediction and feature characterization of plant virus-encoded RNA-silencing suppressors

To counter the host RNA silencing defense mechanism, many plant viruses encode RNA silencing suppressor proteins. These groups of proteins share very low sequence and structural similarities among them, which consequently hamper their annotation using sequence similarity-based search methods. Alternatively the machine learning-based methods can become a suitable choice, but the optimal performance through machine learning-based methods is being affected by various factors such as class imbalance, incomplete learning, selection of inappropriate features, etc. In this paper, we have proposed a novel approach to deal with the class imbalance problem by finding the optimal class distribution for enhancing the prediction accuracy for the RNA silencing suppressors. The optimal class distribution was obtained using different resampling techniques with varying degrees of class distribution starting from natural distribution to ideal distribution, i.e., equal distribution. The experimental results support the fact that optimal class distribution plays an important role to achieve near perfect learning. The best prediction results are obtained with Sequential Minimal Optimization (SMO) learning algorithm. We could achieve a sensitivity of 98.5 %, specificity of 92.6 % with an overall accuracy of 95.3 % on a tenfold cross validation and is further validated using leave one out cross validation test. It was also observed that the machine learning models trained on oversampled training sets using synthetic minority oversampling technique (SMOTE) have relatively performed better than on both randomly undersampled and imbalanced training data sets. Further, we have characterized the important discriminatory sequence features of RNA-silencing suppressors which distinguish these groups of proteins from other protein families.

Improving protein fold recognition and structural class prediction accuracies using physicochemical properties of amino acids

Predicting the three-dimensional (3-D) structure of a protein is an important task in the field of bioinformatics and biological sciences. However, directly predicting the 3-D structure from the primary structure is hard to achieve. Therefore, predicting the fold or structural class of a protein sequence is generally used as an intermediate step in determining the protein's 3-D structure. For protein fold recognition (PFR) and structural class prediction (SCP), two steps are required - feature extraction step and classification step. Feature extraction techniques generally utilize syntactical-based information, evolutionary-based information and physicochemical-based information to extract features. In this study, we explore the importance of utilizing the physicochemical properties of amino acids for improving PFR and SCP accuracies. For this, we propose a Forward Consecutive Search (FCS) scheme which aims to strategically select physicochemical attributes that will supplement the existing feature extraction techniques for PFR and SCP. An exhaustive search is conducted on all the existing 544 physicochemical attributes using the proposed FCS scheme and a subset of physicochemical attributes is identified. Features extracted from these selected attributes are then combined with existing syntactical-based and evolutionary-based features, to show an improvement in the recognition and prediction performance on benchmark datasets.

Prediction of Protein Structural Classes for Low-Similarity Sequences Based on Consensus Sequence and Segmented PSSM

Prediction of protein structural classes for low-similarity sequences is useful for understanding fold patterns, regulation, functions, and interactions of proteins. It is well known that feature extraction is significant to prediction of protein structural class and it mainly uses protein primary sequence, predicted secondary structure sequence, and position-specific scoring matrix (PSSM). Currently, prediction solely based on the PSSM has played a key role in improving the prediction accuracy. In this paper, we propose a novel method called CSP-SegPseP-SegACP by fusing consensus sequence (CS), segmented PsePSSM, and segmented autocovariance transformation (ACT) based on PSSM. Three widely used low-similarity datasets (1189, 25PDB, and 640) are adopted in this paper. Then a 700-dimensional (700D) feature vector is constructed and the dimension is decreased to 224D by using principal component analysis (PCA). To verify the performance of our method, rigorous jackknife cross-validation tests are performed on 1189, 25PDB, and 640 datasets. Comparison of our results with the existing PSSM-based methods demonstrates that our method achieves the favorable and competitive performance. This will offer an important complementary to other PSSM-based methods for prediction of protein structural classes for low-similarity sequences.

A novel Multi-Agent Ada-Boost algorithm for predicting protein structural class with the information of protein secondary structure

Knowledge of the structural class of a given protein is important for understanding its folding patterns. Although a lot of efforts have been made, it still remains a challenging problem for prediction of protein structural class solely from protein sequences. The feature extraction and classification of proteins are the main problems in prediction. In this research, we extended our earlier work regarding these two aspects. In protein feature extraction, we proposed a scheme by calculating the word frequency and word position from sequences of amino acid, reduced amino acid, and secondary structure. For an accurate classification of the structural class of protein, we developed a novel Multi-Agent Ada-Boost (MA-Ada) method by integrating the features of Multi-Agent system into Ada-Boost algorithm. Extensive experiments were taken to test and compare the proposed method using four benchmark datasets in low homology. The results showed classification accuracies of 88.5%, 96.0%, 88.4%, and 85.5%, respectively, which are much better compared with the existing methods. The source code and dataset are available on request.

Identification of human drug targets using machine-learning algorithms

Identification of potential drug targets is a crucial task in the drug-discovery pipeline. Successful identification of candidate drug targets in entire genomes is very useful, and computational prediction methods can speed up this process. In the current work we have developed a sequence-based prediction method for the successful identification and discrimination of human drug target proteins, from human non-drug target proteins. The training features include sequence-based features, such as amino acid composition, amino acid property group composition, and dipeptide composition for generating predictive models. The classification of human drug target proteins presents a classic example of class imbalance. We have addressed this issue by using SMOTE (Synthetic Minority Over-sampling Technique) as a preprocessing step, for balancing the training data with a ratio of 1:1 between drug targets (minority samples) and non-drug targets (majority samples). Using ensemble classification learning method-Rotation Forest and ReliefF feature-selection technique for selecting the optimal subset of salient features, the best model with selected features can achieve 87.1% sensitivity, 83.6% specificity, and 85.3% accuracy, with 0.71 Matthews correlation coefficient (mcc) on a tenfold stratified cross-validation test. The subset of identified optimal features may help in assessing the compositional patterns in human drug targets. For further validation, using a rigorous leave-one-out cross-validation test, the model achieved 88.1% sensitivity, 83.0% specificity, 85.5% accuracy, and 0.712 mcc. The proposed method was tested on a second dataset, for which the current pipeline gave promising results. We suggest that the present approach can be applied successfully as a complementary tool to existing methods for novel drug target prediction.

Characteristics of protein residue-residue contacts and their application in contact prediction

Contact sites between amino acids characterize important structural features of a protein. We investigated characteristics of contact sites in a representative set of proteins and their relations between protein class or topology. For this purpose, we used a non-redundant set of 5872 protein domains, identically categorized by CATH and SCOP databases. The proteins represented alpha, beta, and alpha+beta classes. Contact maps of protein structures were obtained for a selected set of physical distances in the main backbone and separations in protein sequences. For each set a dependency between contact degree and distance parameters was quantified. We indicated residues forming contact sites most frequently and unique amino acid pairs which created contact sites most often within each structural class. Contact characteristics of specific topologies were compared to the characteristics of their protein classes showing protein groups with a distinguished contact characteristic. We showed that our results could be used to improve the performance of recent top contact predictor — direct coupling analysis. Our work provides values of contact site propensities that can be involved in bioinformatic databases.

Discriminating protein structure classes by incorporating Pseudo Average Chemical Shift to Chou's general PseAAC and Support Vector Machine

Proteins control all biological functions in living species. Protein structure is comprised of four major classes including all-α class, all-β class, α+β, and α/β. Each class performs different function according to their nature. Owing to the large exploration of protein sequences in the databanks, the identification of protein structure classes is difficult through conventional methods with respect to cost and time. Looking at the importance of protein structure classes, it is thus highly desirable to develop a computational model for discriminating protein structure classes with high accuracy. For this purpose, we propose a silco method by incorporating Pseudo Average Chemical Shift and Support Vector Machine. Two feature extraction schemes namely Pseudo Amino Acid Composition and Pseudo Average Chemical Shift are used to explore valuable information from protein sequences. The performance of the proposed model is assessed using four benchmark datasets 25PDB, 1189, 640 and 399 employing jackknife test. The success rates of the proposed model are 84.2%, 85.0%, 86.4%, and 89.2%, respectively on the four datasets. The empirical results reveal that the performance of our proposed model compared to existing models is promising in the literature so far and might be useful for future research.

Protein sub-nuclear localization prediction using SVM and Pfam domain information

The nucleus is the largest and the highly organized organelle of eukaryotic cells. Within nucleus exist a number of pseudo-compartments, which are not separated by any membrane, yet each of them contains only a specific set of proteins. Understanding protein sub-nuclear localization can hence be an important step towards understanding biological functions of the nucleus. Here we have described a method, SubNucPred developed by us for predicting the sub-nuclear localization of proteins. This method predicts protein localization for 10 different sub-nuclear locations sequentially by combining presence or absence of unique Pfam domain and amino acid composition based SVM model. The prediction accuracy during leave-one-out cross-validation for centromeric proteins was 85.05%, for chromosomal proteins 76.85%, for nuclear speckle proteins 81.27%, for nucleolar proteins 81.79%, for nuclear envelope proteins 79.37%, for nuclear matrix proteins 77.78%, for nucleoplasm proteins 76.98%, for nuclear pore complex proteins 88.89%, for PML body proteins 75.40% and for telomeric proteins it was 83.33%. Comparison with other reported methods showed that SubNucPred performs better than existing methods. A web-server for predicting protein sub-nuclear localization named SubNucPred has been established at http://14.139.227.92/mkumar/subnucpred/. Standalone version of SubNucPred can also be downloaded from the web-server.

Predict protein structural class for low-similarity sequences by evolutionary difference information into the general form of Chou's pseudo amino acid composition

Knowledge of protein structural class plays an important role in characterizing the overall folding type of a given protein. At present, it is still a challenge to extract sequence information solely using protein sequence for protein structural class prediction with low similarity sequence in the current computational biology. In this study, a novel sequence representation method is proposed based on position specific scoring matrix for protein structural class prediction. By defined evolutionary difference formula, varying length proteins are expressed as uniform dimensional vectors, which can represent evolutionary difference information between the adjacent residues of a given protein. To perform and evaluate the proposed method, support vector machine and jackknife tests are employed on three widely used datasets, 25PDB, 1189 and 640 datasets with sequence similarity lower than 25%, 40% and 25%, respectively. Comparison of our results with the previous methods shows that our method may provide a promising method to predict protein structural class especially for low-similarity sequences.

Protein fold recognition using geometric kernel data fusion

Motivation: Various approaches based on features extracted from protein sequences and often machine learning methods have been used in the prediction of protein folds. Finding an efficient technique for integrating these different protein features has received increasing attention. In particular, kernel methods are an interesting class of techniques for integrating heterogeneous data. Various methods have been proposed to fuse multiple kernels. Most techniques for multiple kernel learning focus on learning a convex linear combination of base kernels. In addition to the limitation of linear combinations, working with such approaches could cause a loss of potentially useful information.

Results: We design several techniques to combine kernel matrices by taking more involved, geometry inspired means of these matrices instead of convex linear combinations. We consider various sequence-based protein features including information extracted directly from position-specific scoring matrices and local sequence alignment. We evaluate our methods for classification on the SCOP PDB-40D benchmark dataset for protein fold recognition. The best overall accuracy on the protein fold recognition test set obtained by our methods is ∼86.7%. This is an improvement over the results of the best existing approach. Moreover, our computational model has been developed by incorporating the functional domain composition of proteins through a hybridization model. It is observed that by using our proposed hybridization model, the protein fold recognition accuracy is further improved to 89.30%. Furthermore, we investigate the performance of our approach on the protein remote homology detection problem by fusing multiple string kernels.

Availability and implementation: The MATLAB code used for our proposed geometric kernel fusion frameworks are publicly available at http://people.cs.kuleuven.be/∼raf.vandebril/homepage/software/geomean.php?menu=5/

Contact:pooyapaydar@gmail.com or yves.moreau@esat.kuleuven.be

Supplementary information:Supplementary data are available at Bioinformatics online.

PalmPred: an SVM based palmitoylation prediction method using sequence profile information

Protein palmitoylation is the covalent attachment of the 16-carbon fatty acid palmitate to a cysteine residue. It is the most common acylation of protein and occurs only in eukaryotes. Palmitoylation plays an important role in the regulation of protein subcellular localization, stability, translocation to lipid rafts and many other protein functions. Hence, the accurate prediction of palmitoylation site(s) can help in understanding the molecular mechanism of palmitoylation and also in designing various related experiments. Here we present a novel in silico predictor called ‘PalmPred’ to identify palmitoylation sites from protein sequence information using a support vector machine model. The best performance of PalmPred was obtained by incorporating sequence conservation features of peptide of window size 11 using a leave-one-out approach. It helped in achieving an accuracy of 91.98%, sensitivity of 79.23%, specificity of 94.30%, and Matthews Correlation Coefficient of 0.71. PalmPred outperformed existing palmitoylation site prediction methods – IFS-Palm and WAP-Palm on an independent dataset. Based on these measures it can be anticipated that PalmPred will be helpful in identifying candidate palmitoylation sites. All the source datasets, standalone and web-server are available at http://14.139.227.92/mkumar/palmpred/.

Protein fold recognition based on functional domain composition

Recognition of protein fold types is an important step in protein structure and function predictions and is also an important method in protein sequence-structure research. Protein fold type reflects the topological pattern of the structure's core. Now there are three methods of protein structure prediction, comparative modeling, fold recognition and de novo prediction. Since comparative modeling is limited by sequence similarity and there is too much workload in de novo prediction, fold recognition has the greatest potential. In order to improve recognition accuracy, a recognition method based on functional domain composition is proposed in this paper. This article focuses on the 124 fold types which have more than 2 samples in LIFCA database. We apply the functional domain composition to predict the fold types of a protein or a domain. In order to evaluate our method and its sensibility to the samples involving SCOP family divided, we tested our results from different aspects. The average sensitivity, specificity and Matthew's correlation coefficient (MCC) of the 124 fold types were found to be 94.58%, 99.96% and 0.91, respectively. Our results indicate that the functional domain composition method is a very promising method for protein fold recognition. And though based on simple classification rules, LIFCA database can grasp the functional features of different proteins, reflecting the corresponding relation between protein structure and function.

High-accuracy prediction of protein structural classes using PseAA structural properties and secondary structural patterns

Since introduction of PseAAs and functional domains, promising results have been achieved in protein structural class predication, but some challenges still exist in the representation of the PseAA structural correlation and structural domains. This paper proposed a high-accuracy prediction method using novel PseAA structural properties and secondary structural patterns, reflecting the long-range and local structural properties of the PseAAs and certain compact structural domains. The proposed prediction method was tested against the competing prediction methods with four experiments. The experiment results indicate that the proposed method achieved the best performance. Its overall accuracies for datasets 25 PDB, D640, FC699 and 1189 are 88.8%, 90.9%, 96.4% and 87.4%, which are 4.5%, 7.6%, 2% and 3.9% higher than the existing best-performing method. This understanding can be used to guide development of more powerful methods for protein structural class prediction. The software and supplement material are freely available at http://bioinfo.zstu.edu.cn/PseAA-SSP.

Prediction of protein structure classes using hybrid space of multi-profile Bayes and bi-gram probability feature spaces

Proteins are the executants of biological functions in living organisms. Comprehension of protein structure is a challenging problem in the era of proteomics, computational biology, and bioinformatics because of its pivotal role in protein folding patterns. Owing to the large exploration of protein sequences in protein databanks and intricacy of protein structures, experimental and theoretical methods are insufficient for prediction of protein structure classes. Therefore, it is highly desirable to develop an accurate, reliable, and high throughput computational model to predict protein structure classes correctly from polygenetic sequences. In this regard, we propose a promising model employing hybrid descriptor space in conjunction with optimized evidence-theoretic K-nearest neighbor algorithm. Hybrid space is the composition of two descriptor spaces including Multi-profile Bayes and bi-gram probability. In order to enhance the generalization power of the classifier, we have selected high discriminative descriptors from the hybrid space using particle swarm optimization, a well-known evolutionary feature selection technique. Performance evaluation of the proposed model is performed using the jackknife test on three low similarity benchmark datasets including 25PDB, 1189, and 640. The success rates of the proposed model are 87.0%, 86.6%, and 88.4%, respectively on the three benchmark datasets. The comparative analysis exhibits that our proposed model has yielded promising results compared to the existing methods in the literature. In addition, our proposed prediction system might be helpful in future research particularly in cases where the major focus of research is on low similarity datasets.

TFpredict and SABINE: sequence-based prediction of structural and functional characteristics of transcription factors

One of the key mechanisms of transcriptional control are the specific connections between transcription factors (TF) and cis-regulatory elements in gene promoters. The elucidation of these specific protein-DNA interactions is crucial to gain insights into the complex regulatory mechanisms and networks underlying the adaptation of organisms to dynamically changing environmental conditions. As experimental techniques for determining TF binding sites are expensive and mostly performed for selected TFs only, accurate computational approaches are needed to analyze transcriptional regulation in eukaryotes on a genome-wide level. We implemented a four-step classification workflow which for a given protein sequence (1) discriminates TFs from other proteins, (2) determines the structural superclass of TFs, (3) identifies the DNA-binding domains of TFs and (4) predicts their cis-acting DNA motif. While existing tools were extended and adapted for performing the latter two prediction steps, the first two steps are based on a novel numeric sequence representation which allows for combining existing knowledge from a BLAST scan with robust machine learning-based classification. By evaluation on a set of experimentally confirmed TFs and non-TFs, we demonstrate that our new protein sequence representation facilitates more reliable identification and structural classification of TFs than previously proposed sequence-derived features. The algorithms underlying our proposed methodology are implemented in the two complementary tools TFpredict and SABINE. The online and stand-alone versions of TFpredict and SABINE are freely available to academics at http://www.cogsys.cs.uni-tuebingen.de/software/TFpredict/ and http://www.cogsys.cs.uni-tuebingen.de/software/SABINE/.

An approach for identifying cytokines based on a novel ensemble classifier

Biology is meaningful and important to identify cytokines and investigate their various functions and biochemical mechanisms. However, several issues remain, including the large scale of benchmark datasets, serious imbalance of data, and discovery of new gene families. In this paper, we employ the machine learning approach based on a novel ensemble classifier to predict cytokines. We directly selected amino acids sequences as research objects. First, we pretreated the benchmark data accurately. Next, we analyzed the physicochemical properties and distribution of whole amino acids and then extracted a group of 120-dimensional (120D) valid features to represent sequences. Third, in the view of the serious imbalance in benchmark datasets, we utilized a sampling approach based on the synthetic minority oversampling technique algorithm and K-means clustering undersampling algorithm to rebuild the training set. Finally, we built a library for dynamic selection and circulating combination based on clustering (LibD3C) and employed the new training set to realize cytokine classification. Experiments showed that the geometric mean of sensitivity and specificity obtained through our approach is as high as 93.3%, which proves that our approach is effective for identifying cytokines.

Comparison study on statistical features of predicted secondary structures for protein structural class prediction: From content to position

BACKGROUND

Many content-based statistical features of secondary structural elements (CBF-PSSEs) have been proposed and achieved promising results in protein structural class prediction, but until now position distribution of the successive occurrences of an element in predicted secondary structure sequences hasn't been used. It is necessary to extract some appropriate position-based features of the secondary structural elements for prediction task.

RESULTS

We proposed some position-based features of predicted secondary structural elements (PBF-PSSEs) and assessed their intrinsic ability relative to the available CBF-PSSEs, which not only offers a systematic and quantitative experimental assessment of these statistical features, but also naturally complements the available comparison of the CBF-PSSEs. We also analyzed the performance of the CBF-PSSEs combined with the PBF-PSSE and further constructed a new combined feature set, PBF11CBF-PSSE. Based on these experiments, novel valuable guidelines for the use of PBF-PSSEs and CBF-PSSEs were obtained.

CONCLUSIONS

PBF-PSSEs and CBF-PSSEs have a compelling impact on protein structural class prediction. When combining with the PBF-PSSE, most of the CBF-PSSEs get a great improvement over the prediction accuracies, so the PBF-PSSEs and the CBF-PSSEs have to work closely so as to make significant and complementary contributions to protein structural class prediction. Besides, the proposed PBF-PSSE's performance is extremely sensitive to the choice of parameter k. In summary, our quantitative analysis verifies that exploring the position information of predicted secondary structural elements is a promising way to improve the abilities of protein structural class prediction.

Accurate prediction of protein structural class

Because of the increasing gap between the data from sequencing and structural genomics, the accurate prediction of the structural class of a protein domain solely from the primary sequence has remained a challenging problem in structural biology. Traditional sequence-based predictors generally select several sequence features and then feed them directly into a classification program to identify the structural class. The current best sequence-based predictor achieved an overall accuracy of 74.1% when tested on a widely used, non-homologous benchmark dataset 25PDB. In the present work, we built a multiple linear regression (MLR) model to convert the 440-dimensional (440D) sequence feature vector extracted from the Position Specific Scoring Matrix (PSSM) of a protein domain to a 4-dimensinal (4D) structural feature vector, which could then be used to predict the four major structural classes. We performed 10-fold cross-validation and jackknife tests of the method on a large non-homologous dataset containing 8,244 domains distributed among the four major classes. The performance of our approach outperformed all of the existing sequence-based methods and had an overall accuracy of 83.1%, which is even higher than the results of those predicted secondary structure-based methods.

Prediction of protein domain with mRMR feature selection and analysis

The domains are the structural and functional units of proteins. With the avalanche of protein sequences generated in the postgenomic age, it is highly desired to develop effective methods for predicting the protein domains according to the sequences information alone, so as to facilitate the structure prediction of proteins and speed up their functional annotation. However, although many efforts have been made in this regard, prediction of protein domains from the sequence information still remains a challenging and elusive problem. Here, a new method was developed by combing the techniques of RF (random forest), mRMR (maximum relevance minimum redundancy), and IFS (incremental feature selection), as well as by incorporating the features of physicochemical and biochemical properties, sequence conservation, residual disorder, secondary structure, and solvent accessibility. The overall success rate achieved by the new method on an independent dataset was around 73%, which was about 28–40% higher than those by the existing method on the same benchmark dataset. Furthermore, it was revealed by an in-depth analysis that the features of evolution, codon diversity, electrostatic charge, and disorder played more important roles than the others in predicting protein domains, quite consistent with experimental observations. It is anticipated that the new method may become a high-throughput tool in annotating protein domains, or may, at the very least, play a complementary role to the existing domain prediction methods, and that the findings about the key features with high impacts to the domain prediction might provide useful insights or clues for further experimental investigations in this area. Finally, it has not escaped our notice that the current approach can also be utilized to study protein signal peptides, B-cell epitopes, HIV protease cleavage sites, among many other important topics in protein science and biomedicine.

Predicting sumoylation site by feature selection method

The small ubiquitin-like modifier (SUMO) proteins are a kind of proteins that can be attached to a series of proteins. The sumoylation of protein is an important posttranslational modification. Thus, the prediction of the sumoylation site of a given protein is significant. Here we employed a combined method to perform this task. We predicted the sumoylation site of a protein by a two-staged procedure. At the first stage, whether a protein would be sumoylated was predicted; whereas at the second stage, the sumoylation sites of the protein were predicted if it was determined to be modified by SUMO at the first stage. At the first stage, we encoded a protein with protein families (PFAM) and trained the predictor with nearest network algorithm (NNA); at the second stage, we encoded nonapeptides (peptides that contain nine residues) of the protein containing the lysine residues, with Amino Acid Index, and trained the predictor with NNA. The predictor was tested by the k-fold cross-validation method. The highest accuracy of the second-staged predictor was 99.55% when 12 features were incorporated in the predictor. The corresponding Matthews Correlation Coefficient was 0.7952. These results indicate that the method is a promising tool to predict the sumoylation site of a protein. At last, the features used in the predictor are discussed. The software is available at request.

Accurate prediction of protein structural class using auto covariance transformation of PSI-BLAST profiles

Computational prediction of protein structural class based solely on sequence data remains a challenging problem in protein science. Existing methods differ in the protein sequence representation models and prediction engines adopted. In this study, a powerful feature extraction method, which combines position-specific score matrix (PSSM) with auto covariance (AC) transformation, is introduced. Thus, a sample protein is represented by a series of discrete components, which could partially incorporate the long-range sequence order information and evolutionary information reflected from the PSI-BLAST profile. To verify the performance of our method, jackknife cross-validation tests are performed on four widely used benchmark datasets. Comparison of our results with existing methods shows that our method provides the state-of-the-art performance for structural class prediction. A Web server that implements the proposed method is freely available at http://202.194.133.5/xinxi/AAC_PSSM_AC/index.htm.

Multitask learning for protein subcellular location prediction

Protein subcellular localization is concerned with predicting the location of a protein within a cell using computational methods. The location information can indicate key functionalities of proteins. Thus, accurate prediction of subcellular localizations of proteins can help the prediction of protein functions and genome annotations, as well as the identification of drug targets. Machine learning methods such as Support Vector Machines (SVMs) have been used in the past for the problem of protein subcellular localization, but have been shown to suffer from a lack of annotated training data in each species under study. To overcome this data sparsity problem, we observe that because some of the organisms may be related to each other, there may be some commonalities across different organisms that can be discovered and used to help boost the data in each localization task. In this paper, we formulate protein subcellular localization problem as one of multitask learning across different organisms. We adapt and compare two specializations of the multitask learning algorithms on 20 different organisms. Our experimental results show that multitask learning performs much better than the traditional single-task methods. Among the different multitask learning methods, we found that the multitask kernels and supertype kernels under multitask learning that share parameters perform slightly better than multitask learning by sharing latent features. The most significant improvement in terms of localization accuracy is about 25 percent. We find that if the organisms are very different or are remotely related from a biological point of view, then jointly training the multiple models cannot lead to significant improvement. However, if they are closely related biologically, the multitask learning can do much better than individual learning.