Protein and Peptide Letters - Volume 17, Issue 9, 2010

We are very pleased to offer this special issue to the readers of Protein and Peptide Letters by selecting the candidate papers from the 2009 International Conference on Intelligent Computing (ICIC), held on September 16-19, 2009 in Ulsan, Korea. Eleven papers representing less than five percent of all eligible papers accepted at the ICIC2009 are selected for inclusion in this special issue. In recent years, we have witnessed intelligent computing techniques, such as artificial intelligence, machine learning, feature selection, ensembles and others being dedicated to various research aspects of bioinformatics, chemoinformatics, computational biology, system biology, etc. Meanwhile, intelligent computing research has been enriched by the development of more solid mathematical frameworks, elaborating more efficient and powerful algorithms. More importantly, it has been driven by its application to many amazing research fields, such as bioinformatics. Currently, intelligent computing techniques for protein bioinformatics are being used to encode protein and peptide biological information, extract biological features, recognize biological patterns, mine and comprehend biological data, build models of biological systems and processes, and automatically form theories from the unprecedentedly vast experimental biological data on protein sequence, genetics, structure, interactions and functions, etc. Its main objective is to find the rule and useful biological information and extract biological knowledge from limited observation examples that cannot be obtained using classical biological methods and theories. It extends the rule to predict and infer protein's structure, function, interaction networks, and other significant aspects of protein and peptide science. Hence, the intelligent computing technique, an in silico method, is supplementary to conventional experimental methods. This special issue includes eleven papers on how to use intelligent computing techniques to solve problems in protein bioinformatics. Four papers in this issue focus on computational algorithms application in protein-protein interaction (PPI) prediction and PPI network analysis. Xia, et al. discuss the existing computational methods for PPI prediction from protein genetics, sequence and structure information, etc. Yang et al. predict protein-protein interactions using local descriptors that account for the interactions between residues in both continuous and discontinuous regions of a protein sequence and using a k-nearest neighbor classification system. Wang et al., infer PPI with a hybrid Genetic Algorithm /Support Vector Machine method by representing the protein with its domains, where they consider the effects of duplication, combination transformation of the domain composition using genetic algorithm (GA) and discover the optimal transformation with a support vector machine (SVM) classifier. Lee et al. provide a fast and adaptive approach to revel the highly evolutionarily conserved motif mode of a yeast protein interaction network through intelligent agent-based distributed computing method. The next two papers present intelligent computing methods on protein interaction site predictions. Han, et al. compute the interaction propensity of three consecutive amino acids (called amino acid triplet or triple amino acids) from the structure data of protein-RNA complexes and predict RNA-binding sites in proteins using SVM with the interaction propensity, which are proved to be more effective than other amino acid biochemical properties. Wang, et al. use a radial basis function neural network (RBFNN) ensemble model to predict protein interaction sites in heterocomplexes by classifing protein surface residues into interaction sites or non-interaction sites and judging the prediction output by majority voting. Another two papers focus on protein supersecondary structure prediction. Specifically, Xia et al. propose a two-stage SVM classifier with new physicochemical and structural properties-based coding schemes to predict π-turns that are irregular secondary structure elements consisting of short backbone fragments (six-amino-acid residues) where the backbone reverses its overall direction and play an important role in proteins from both the structural and functional points of view. Xia et al. further propose a SVM ensemble with majority voting strategy and new feature representation scheme based on auto covariance combining with protein secondary structure and residue conformation propensity to predict β-hairpins in proteins more accurately. The last three papers focus on applying feature selection method, iterative partitioning technique and simple linear regression predictor to TAP binding peptides, protein repeats, and peptide drift time in ion mobility-mass spectrometry, respectively. More specifically, Li, et al. present a new feature selection approach to predict and analyze TAP-peptide binding specificity using Forward Attribute Reduction based on Neighborhood Model (FARNeM). Zhang, et al. propose weight protein sequence with respective probabilities of occurrence at each position and present the iterative partitioning technique to define and locate the loose repeats and the strict repeats with the weighted sequence. Wang, et al. introduce a method for predicting peptide drift time in ion mobility-mass spectrometry (IMMS) with a numeric descriptor, i.e. molecular weight for peptide representation and a simple linear regression predictor for peptides drift time prediction....

Protein-protein interactions (PPIs) are key components of most cellular processes, so identification of PPIs is at the heart of functional genomics. A number of experimental techniques have been developed to discover the PPI networks of several organisms. However, the accuracy and coverage of these techniques have proven to be limited. Therefore, it is important to develop computational methods to assist in the design and validation of experimental studies and for the prediction of interaction partners. Here, we provide a critical overview of existing computational methods including genomic context method, structure-based method, domain-based method and sequence-based method. While an exhaustive list of methods is not presented, we analyze the relative strengths and weaknesses for each of the methods discussed, as well as a broader perspective on computational techniques for determining PPIs. In addition to algorithms for interaction prediction, description of many useful databases pertaining to PPIs is also provided.

Identifying protein-protein interaction is crucial for understanding the biological systems and processes, as well as mutant design. This paper proposes a novel hybrid Genetic Algorithm/Support Vector Machine (GA/SVM) method to predict the interactions between proteins intermediated by the protein-domain relations. A protein domain is a structural and/or functional unit of the protein. Every protein can be characterized by a distinct domain or a sequential combination of multiple domains. In our method, the protein was first represented by its domains where the effects of domain duplication were also considered. Transformation of the domain composition was taken to simulate the combination of different domains using genetic algorithm (GA). The optimal transformation was discovered using a predictor constructed by a support vector machines (SVM) method. Compared with random predictor, the prediction performance of our method is more effective and efficient with 0.85 sensitivity, 0.90 specificity and 0.88 accuracy.

With a huge amount of protein sequence data, the computational method for protein-protein interaction (PPI) prediction using only the protein sequences information have drawn increasing interest. In this article, we propose a sequence- based method based on a novel representation of local protein sequence descriptors. Local descriptors account for the interactions between residues in both continuous and discontinuous regions of a protein sequence, so this method enables us to extract more PPI information from the sequence. A series of elaborate experiments are performed to optimize the prediction model by varying the parameter k and the distance measuring function of the k-nearest neighbors learning system and the ways of coding a protein pair. When performed on the PPI data of Saccharomyces cerevisiae, the method achieved 86.15% prediction accuracy with 81.03% sensitivity at the precision of 90.24%. An independent data set of 986 Escherichia coli PPIs was used to evaluate this prediction model and the prediction accuracy is 73.02%. Given the complex nature of PPIs, the performance of our method is promising, and it can be a helpful supplement for PPIs prediction.

In the yeast protein-protein interaction network, motif mode, a collection of motifs of special combinations of protein nodes annotated by the molecular function terms of the Gene Ontology, has revealed differences in the conservation constraints within the same topology. In this study, by employing an intelligent agent-based distributed computing method, we are able to discover motif modes in a fast and adaptive manner. Moreover, by focusing on the highly evolutionarily conserved motif modes belonging to the same biological function, we find a large downshift in the distance between nodes belonging to the same motif mode compared with the whole, suggesting that nodes with the same motif mode tend to congregate in a network. Several motif modes with a high conservation of the motif constituents were revealed, but from a new perspective, including that with a three-node motif mode engaged in the protein fate and that with three four-node motif modes involved in the genome maintenance, cellular organization, and transcription. The network motif modes discovered from this method can be linked to the wealth of biological data which require further elucidation with regard to biological functions.

In protein-RNA interactions, amino acids often exhibit different preferences for its RNA partners with different neighbor amino acids. Hence, the interaction propensity of an amino acid can be better assessed by considering neighbors of the amino acid than examining the amino acid alone. In this study, we computed the interaction propensity of three consecutive amino acids (called amino acid triplet or triple amino acids) from the rigorous analysis of the recent structure data of protein-RNA complexes. We used the interaction propensity to predict RNA-binding sites in protein sequences with a support vector machine (SVM) classifier, and observed that the interaction propensities of amino acid triplets are more effective than other biochemical properties of amino acids for predicting RNA-binding sites in proteins. Experimental results with non-redundant 134 protein sequences showed that the SVM classifier achieved a sensitivity of 77% and specificity of 76% and that the three-residue interaction propensity resulted in a better performance than single- or fiveresidue interaction propensities. Comparison of the SVM classifier with RNABindR and BindN demonstrated that it outperforms the other two methods in the net prediction and correlation coefficient. Our SVM classifier can also be used to predict protein-binding nucleotides in RNA sequences.

Prediction of protein-protein interaction sites can guide the structural elucidation of protein complexes. We propose a novel method using a radial basis function neural network (RBFNN) ensemble model for the prediction of protein interaction sites in heterocomplexes. We classified protein surface residues into interaction sites or non-interaction sites based on the RBFNNs trained on different datasets, then judged a prediction to be the final output. Only information of evolutionary conservation and spatial sequence profile are used in this ensemble predictor to describe the protein sites. A non-redundant data set of heterodimers used is consisted of 69 protein chains, in which 10329 surface residues can be found. The efficiency and the effectiveness of our proposed approach can be validated by a better performance such as the accuracy of 0.689, the sensitivity of 66.6% and the specificity of 67.6%.

π-turns are irregular secondary structure elements consisting of short backbone fragments (six-amino-acid residues) where the backbone reverses its overall direction. They play an important role in proteins from both the structural and functional points of view. Recently, some methods have been proposed to predict π-turns. In this study, a new method of π-turn prediction that uses a two-stage classification scheme is proposed based on support vector machine. In addition, different from previous methods, new coding schemes based on the physicochemical properties and the structural properties of proteins are adopted. Seven-fold cross validation based on a dataset of 640 non-homologue protein chains is used to evaluate the performance of our method. The experiment results show our method can yield a promising performance, which confirms the effectiveness of the proposed approach.

In this study, we propose a new method to predict β-Hairpins in proteins and its evaluation based on the support vector machine. Different from previous methods, new feature representation scheme based on auto covariance is adopted. We also investigate two structure properties of proteins (protein secondary structure and residue conformation propensity), and examine their effects on prediction. Moreover, we employ an ensemble classifier approach based on the majority voting to improve prediction accuracy on hairpins. Experimental results on a dataset of 1926 protein chains show that our approach outperforms those previously published in the literature, which demonstrates the effectiveness of the proposed method.

Peptide fragments that serve as the cytotoxic T lymphocyte (CTL) epitopes are processed from antigens by the proteasome and then are transported to the endoplasmic reticulum through transporter associated with antigen processing (TAP) before being loaded onto the MHC class I molecule. Here, we studied TAP specificity by a neighborhood rough set (NRS) model based feature selection and prioritization method. By means of binary, amino acid properties, and binary plus properties of amino acids encoding, respectively, we adopted NRS based feature selection method to select multiple optimal feature sets for TAP binding peptides binary classification. The features in these optimal sets were ranked according to their occurrence frequency. Results show that the NRS is effective for prediction improvement and analysis of the specificity of TAP transporter. The proposed method can be used as a tool for predicting TAP binding peptides and be useful for subunit vaccine rational design and related bioinformatics cases.

A weighted sequence is a string in which a set of characters may appear at each position with respective probabilities of occurrence. Weighted sequences are able to summarize poorly defined short sequences, as well as the profiles of protein families and complete chromosome sequences. Thus it is of biological and theoretical significance to design powerful algorithms on weighted sequences. A common task is to identify repetitive motifs in weighted sequences, with presence probability not less than a given threshold. We define two types of repeats in weighted sequences, called the loose repeats and the strict repeats, respectively, and then attempt to locate these repeats. Using an iterative partitioning technique, we present algorithms for computing all the loose repeats and strict repeats of every length, respectively. Each solution costs O(n2)time.

A computational model is introduced for predicting peptide drift time in ion mobility-mass spectrometry (IMMS). Each peptide was represented using a numeric descriptor: molecular weight. A simple linear regression predictor was constructed for peptides drift time prediction. Three datasets with different charge state assignments were used for the model training and testing. The dataset one contains 212 singly charged peptides, dataset two has 306 doubly charged peptides, and dataset three contains 77 triply charged peptides. Our proposed method achieved a prediction accuracy of 86.3%, 72.6%, and 59.7% for the dataset one, two and three, respectively. Peptide drift time prediction in IMMS will improve the confidence of peptide identifications by limiting the peptide search space during MS/MS database searching and therefore, reducing false discovery rate (FDR) of protein identification.

A Surface Plasmon Resonance Imaging (SPRI) sensor has been developed for highly selective determination of cathepsin D (Cat D) or/and E (Cat E). The sensor contains immobilised pepstatin A, which binds aspartyl proteases from solution. Pepstatin A activated with N-Hydroxysuccinimide (NHS) and N-Ethyl-N'-(3-dimethylaminopropyl) carbodiimide (EDC) was immobilized on an amine-modified gold surface. Cysteamine was used for modification of the gold surface. Pepstatin A concentration and pH of interaction were optimised. A concentration of pepstatin equal to 0.5 μg mL-1 and a pH of 3.75 were selected as optimal.The sensor's dynamic response range is between 0.25 and 1.0 ng mL-1, and the detection limit is 0.12 ng mL-1. However, the sensor cannot distinguish between Cat D and Cat E. In order to demonstrate the sensor's potential, Cat E was determined in human red blood cells, Cat D in human saliva, as well as total concentration of Cat D and Cat E in human nasal polyps.

The lens protein, γB-crystallin, precipitates during cataract formation. As a recombinant protein, in aqueous solution, γB aggregates and precipitates upon heating, cooling, exposure to ultraviolet light, or refolding from a denatured state. We have studied soluble γB crystallin, as well as each of the above aggregated forms, to determine whether γB's polypeptide chain is differently organized in each form. For this purpose, we used : (a) Fourier Transform Infra Red (FTIR) spectroscopy in the horizontal attenuated total reflectance (HATR) mode, to examine changes in secondary structural content, and (b) transmission electron microscopy (TEM) to examine gross morphological differences. The peak of the γB FTIR amide I band shifts from ~1633 cm-1 to ~1618 cm-1 in heat-, UV- and refolding-induced γB precipitates, indicating that narrow beta sheets with fewer strands and higher strand twist angles are becoming reorganized into wider, more planar sheets containing larger numbers of shorter strands, with smaller twist angles. In contrast, in cold-induced precipitates, a loss of anti-parallel beta sheet content is observed. This difference is partly explained by the differential effects of temperature on different non-covalent interactions stabilizing protein structures. The native beta sheet content of γB crystallin (~50%) is raised in heat- (~60%) and refolding-induced (~58%) precipitates, but lowered in cold- (~41%), and UV-induced (~44%) precipitates. Cold precipitates also display ~26% helical content. All four aggregates have distinctively different morphological characteristics; this appears to be in keeping with their distinctively different secondary structural contents.

It has generally been accepted that an increase in protein stability is proportional to the increase in hydrophobicity. When a cavity is created by large-to-small substitutions of amino acid residues in protein cores, protein stability decreases 5.3 kJ/mol per single methyl(ene) group removal. In contrast, many reported cavity-filling mutations either failed to increase stability or produced marginal increases in stability; even in successful cases, the increase in stability was much lower than expected from the cost of single methyl(ene) group removal in cavity-creating mutations. Previously it was found that some cavity-filling mutant proteins at Ala52 in E. coli RNase HI increased stability, but decreased activity and they did not increase the stability to the degree expected by the hydrophobic effect alone. The present study attempted to structurally analyze these variant proteins, and it was found that substitutions have little effect on the overall fold but cause conformational strains with the neighboring residues. The present results and literature on cavity-creating/- filling variants provide insight into protein architecture, indicating that natural protein cores are able to accommodate larger side-chain residue by substitution; in other words, excess-packing may not be chosen in natural selection.

Bla g 6, a German cockroach allergen, shows homology to muscle protein troponin C. It contains four calcium-binding domains at 20-30, 56-67, 96-107, and 132-143 amino acid (aa) residues, and its immunoglobulin E (IgE) reactivity is dependent upon calcium ion level. However, the IgE binding epitopes of Bla g 6 have not been investigated. This study aimed to analyze the IgE binding epitopes from the five peptide fragments of Bla g 6. The full-length of three Bla g 6 isoallergens (Bla g 6.0101, Bla g 6.0201, and Bla g 6.0301) and five peptide fragments (P1: aa 1-111, P2: aa 1-95, P3: aa 33-111, P4: aa 80-151, and P5: aa 33-151) of Bla g 6.0101 were generated by polymerase chain reaction (PCR) and expressed in Escherichia coli. Enzyme-linked immunosorbent assay (ELISA) was performed on 24 patients' sera that adjusted the final concentration 10 mM of CaCl2 to determine the IgE activities of Bla g 6. Eight sera (33.3%), 9 sera (37.5%), and 11 sera (45.8%) showed IgE reactivity to Bla g 6.0101, Bla g 6.0201, and Bla g 6.0301, respectively. Among the sera from the positive IgE reactivity, three patients' sera were selected and the IgE reactivity was measured by ELISA with the five peptide fragments of Bla g 6. Based on IgE responses, one patient's serum exhibited the strongest IgE reactivity. We assumed that the aa between 96-151 residues, including the calcium binding domains III and IV, would be important for IgE binding. These results may provide information that will yield safe diagnostic methods and immunotherapeutics.

There is a growing body of evidence that the formation and accumulation of advanced glycation end products (AGE) have been known to progress under diabetic conditions, thereby being involved in diabetic vascular complications. Further, we, along with others, have recently found AGE could disturb insulin actions in cultured adipocytes and skeletal muscles. However, the pathological role of AGE in insulin resistance in vivo is not fully understood. Therefore, in this study, we examined whether pyridoxamine, an inhibitor of AGE formation could ameliorate insulin resistance in KK-Ay mice, a model animal of obese, type 2 diabetes. Fasting blood glucose, serum levels of insulin and AGE in KK-Ay mice were elevated as the mice got older (from 5 weeks old to 15 weeks old). Serum levels of AGE were positively correlated with insulin (R2=0.3956, P=0.002) in KK-Ay mice. Administration of pyridoxamine dose-dependently decreased fasting insulin levels and improved insulin sensitivity in KK-Ay mice of 10 weeks old, although it did not affect fasting blood glucose levels. Our present study suggests the involvement of AGE in insulin resistance in KK-Ay mice. Inhibition of AGE formation may be a novel therapeutic target for improving insulin resistance in diabetes with obesity.