Darby Tien Hao Chang

Predicting protein-protein interactions in unbalanced data using the primary structure of proteins

BackgroundElucidating protein-protein interactions (PPIs) is essential to constructing protein interaction networks and facilitating our understanding of the general principles of biological systems. Previous studies have revealed that interacting protein pairs can be predicted by their primary structure. Most of these approaches have achieved satisfactory performance on datasets comprising equal number of interacting and non-interacting protein pairs. However, this ratio is highly unbalanced in nature, and these techniques have not been comprehensively evaluated with respect to the effect of the large number of non-interacting pairs in realistic datasets. Moreover, since highly unbalanced distributions usually lead to large datasets, more efficient predictors are desired when handling such challenging tasks.ResultsThis study presents a method for PPI prediction based only on sequence information, which contributes in three aspects. First, we propose a probability-based mechanism for transforming protein sequences into feature vectors. Second, the proposed predictor is designed with an efficient classification algorithm, where the efficiency is essential for handling highly unbalanced datasets. Third, the proposed PPI predictor is assessed with several unbalanced datasets with different positive-to-negative ratios (from 1:1 to 1:15). This analysis provides solid evidence that the degree of dataset imbalance is important to PPI predictors.ConclusionsDealing with data imbalance is a key issue in PPI prediction since there are far fewer interacting protein pairs than non-interacting ones. This article provides a comprehensive study on this issue and develops a practical tool that achieves both good prediction performance and efficiency using only protein sequence information.

Using a kernel density estimation based classifier to predict species-specific microRNA precursors

BackgroundMicroRNAs (miRNAs) are short non-coding RNA molecules participating in post-transcriptional regulation of gene expression. There have been many efforts to discover miRNA precursors (pre-miRNAs) over the years. Recently, ab initio approaches obtain more attention because that they can discover species-specific pre-miRNAs. Most ab initio approaches proposed novel features to characterize RNA molecules. However, there were fewer discussions on the associated classification mechanism in a miRNA predictor.ResultsThis study focuses on the classification algorithm for miRNA prediction. We develop a novel ab initio method, miR-KDE, in which most of the features are collected from previous works. The classification mechanism in miR-KDE is the relaxed variable kernel density estimator (RVKDE) that we have recently proposed. When compared to the famous support vector machine (SVM), RVKDE exploits more local information of the training dataset. MiR-KDE is evaluated using a training set consisted of only human pre-miRNAs to predict a benchmark collected from 40 species. The experimental results show that miR-KDE delivers favorable performance in predicting human pre-miRNAs and has advantages for pre-miRNAs from the genera taxonomically distant to humans.ConclusionWe use a novel classifier of which the characteristic of exploiting local information is particularly suitable to predict species-specific pre-miRNAs. This study also provides a comprehensive analysis from the view of classification mechanism. The good performance of miR-KDE encourages more efforts on the classification methodology as well as the feature extraction in miRNA prediction.

YPA: an integrated repository of promoter features in Saccharomyces cerevisiae

This study presents the Yeast Promoter Atlas (YPA, http://ypa.ee.ncku.edu.tw/ or http://ypa.csbb.ntu.edu.tw/) database, which aims to collect comprehensive promoter features in Saccharomyces cerevisiae. YPA integrates nine kinds of promoter features including promoter sequences, genes’ transcription boundaries—transcription start sites (TSSs), five prime untranslated regions (5′-UTRs) and three prime untranslated regions (3′UTRs), TATA boxes, transcription factor binding sites (TFBSs), nucleosome occupancy, DNA bendability, transcription factor (TF) binding, TF knockout expression and TF–TF physical interaction. YPA is designed to present data in a unified manner as many important observations are revealed only when these promoter features are considered altogether. For example, DNA rigidity can prevent nucleosome packaging, thereby making TFBSs in the rigid DNA regions more accessible to TFs. Integrating nucleosome occupancy, DNA bendability, TF binding, TF knockout expression and TFBS data helps to identify which TFBS is actually functional. In YPA, various promoter features can be accessed in a centralized and organized platform. Researchers can easily view if the TFBSs in an interested promoter are occupied by nucleosomes or located in a rigid DNA segment and know if the expression of the downstream gene responds to the knockout of the corresponding TFs. Compared to other established yeast promoter databases, YPA collects not only TFBSs but also many other promoter features to help biologists study transcriptional regulation.

Real value prediction of protein solvent accessibility using enhanced PSSM features

BackgroundPrediction of protein solvent accessibility, also called accessible surface area (ASA) prediction, is an important step for tertiary structure prediction directly from one-dimensional sequences. Traditionally, predicting solvent accessibility is regarded as either a two- (exposed or buried) or three-state (exposed, intermediate or buried) classification problem. However, the states of solvent accessibility are not well-defined in real protein structures. Thus, a number of methods have been developed to directly predict the real value ASA based on evolutionary information such as position specific scoring matrix (PSSM).ResultsThis study enhances the PSSM-based features for real value ASA prediction by considering the physicochemical properties and solvent propensities of amino acid types. We propose a systematic method for identifying residue groups with respect to protein solvent accessibility. The amino acid columns in the PSSM profile that belong to a certain residue group are merged to generate novel features. Finally, support vector regression (SVR) is adopted to construct a real value ASA predictor. Experimental results demonstrate that the features produced by the proposed selection process are informative for ASA prediction.ConclusionExperimental results based on a widely used benchmark reveal that the proposed method performs best among several of existing packages for performing ASA prediction. Furthermore, the feature selection mechanism incorporated in this study can be applied to other regression problems using the PSSM. The program and data are available from the authors upon request.

Predicting the protein-protein interactions using primary structures with predicted protein surface

BackgroundMany biological functions involve various protein-protein interactions (PPIs). Elucidating such interactions is crucial for understanding general principles of cellular systems. Previous studies have shown the potential of predicting PPIs based on only sequence information. Compared to approaches that require other auxiliary information, these sequence-based approaches can be applied to a broader range of applications.ResultsThis study presents a novel sequence-based method based on the assumption that protein-protein interactions are more related to amino acids at the surface than those at the core. The present method considers surface information and maintains the advantage of relying on only sequence data by including an accessible surface area (ASA) predictor recently proposed by the authors. This study also reports the experiments conducted to evaluate a) the performance of PPI prediction achieved by including the predicted surface and b) the quality of the predicted surface in comparison with the surface obtained from structures. The experimental results show that surface information helps to predict interacting protein pairs. Furthermore, the prediction performance achieved by using the surface estimated with the ASA predictor is close to that using the surface obtained from protein structures.ConclusionThis work presents a sequence-based method that takes into account surface information for predicting PPIs. The proposed procedure of surface identification improves the prediction performance with an F-measure of 5.1%. The extracted surfaces are also valuable in other biomedical applications that require similar information.

Protemot: prediction of protein binding sites with automatically extracted geometrical templates

Geometrical analysis of protein tertiary substructures has been an effective approach employed to predict protein binding sites. This article presents the Protemot web server that carries out prediction of protein binding sites based on the structural templates automatically extracted from the crystal structures of protein–ligand complexes in the PDB (Protein Data Bank). The automatic extraction mechanism is essential for creating and maintaining a comprehensive template library that timely accommodates to the new release of PDB as the number of entries continues to grow rapidly. The design of Protemot is also distinctive by the mechanism employed to expedite the analysis process that matches the tertiary substructures on the contour of the query protein with the templates in the library. This expediting mechanism is essential for providing reasonable response time to the user as the number of entries in the template library continues to grow rapidly due to rapid growth of the number of entries in PDB. This article also reports the experiments conducted to evaluate the prediction power delivered by the Protemot web server. Experimental results show that Protemot can deliver a superior prediction power than a web server based on a manually curated template library with insufficient quantity of entries. Availability: .

A study on the flexibility of enzyme active sites.

BackgroundA common assumption about enzyme active sites is that their structures are highly conserved to specifically distinguish between closely similar compounds. However, with the discovery of distinct enzymes with similar reaction chemistries, more and more studies discussing the structural flexibility of the active site have been conducted.ResultsMost of the existing works on the flexibility of active sites focuses on a set of pre-selected active sites that were already known to be flexible. This study, on the other hand, proposes an analysis framework composed of a new data collecting strategy, a local structure alignment tool and several physicochemical measures derived from the alignments. The method proposed to identify flexible active sites is highly automated and robust so that more extensive studies will be feasible in the future. The experimental results show the proposed method is (a) consistent with previous works based on manually identified flexible active sites and (b) capable of identifying potentially new flexible active sites.ConclusionsThis proposed analysis framework and the former analyses on flexibility have their own advantages and disadvantage, depending on the cause of the flexibility. In this regard, this study proposes an alternative that complements previous studies and helps to construct a more comprehensive view of the flexibility of enzyme active sites.

Predicting microRNA precursors with a generalized Gaussian components based density estimation algorithm

BackgroundMicroRNAs (miRNAs) are short non-coding RNA molecules, which play an important role in post-transcriptional regulation of gene expression. There have been many efforts to discover miRNA precursors (pre-miRNAs) over the years. Recently, ab initio approaches have attracted more attention because they do not depend on homology information and provide broader applications than comparative approaches. Kernel based classifiers such as support vector machine (SVM) are extensively adopted in these ab initio approaches due to the prediction performance they achieved. On the other hand, logic based classifiers such as decision tree, of which the constructed model is interpretable, have attracted less attention.ResultsThis article reports the design of a predictor of pre-miRNAs with a novel kernel based classifier named the generalized Gaussian density estimator (G2DE) based classifier. The G2DE is a kernel based algorithm designed to provide interpretability by utilizing a few but representative kernels for constructing the classification model. The performance of the proposed predictor has been evaluated with 692 human pre-miRNAs and has been compared with two kernel based and two logic based classifiers. The experimental results show that the proposed predictor is capable of achieving prediction performance comparable to those delivered by the prevailing kernel based classification algorithms, while providing the user with an overall picture of the distribution of the data set.ConclusionSoftware predictors that identify pre-miRNAs in genomic sequences have been exploited by biologists to facilitate molecular biology research in recent years. The G2DE employed in this study can deliver prediction accuracy comparable with the state-of-the-art kernel based machine learning algorithms. Furthermore, biologists can obtain valuable insights about the different characteristics of the sequences of pre-miRNAs with the models generated by the G2DE based predictor.

Combining Phylogenetic Profiling-Based and Machine Learning-Based Techniques to Predict Functional Related Proteins

Annotating protein functions and linking proteins with similar functions are important in systems biology. The rapid growth rate of newly sequenced genomes calls for the development of computational methods to help experimental techniques. Phylogenetic profiling (PP) is a method that exploits the evolutionary co-occurrence pattern to identify functional related proteins. However, PP-based methods delivered satisfactory performance only on prokaryotes but not on eukaryotes. This study proposed a two-stage framework to predict protein functional linkages, which successfully enhances a PP-based method with machine learning. The experimental results show that the proposed two-stage framework achieved the best overall performance in comparison with three PP-based methods.

AH-DB: collecting protein structure pairs before and after binding

This work presents the Apo–Holo DataBase (AH-DB, http://ahdb.ee.ncku.edu.tw/ and http://ahdb.csbb.ntu.edu.tw/), which provides corresponding pairs of protein structures before and after binding. Conformational transitions are commonly observed in various protein interactions that are involved in important biological functions. For example, copper–zinc superoxide dismutase (SOD1), which destroys free superoxide radicals in the body, undergoes a large conformational transition from an ‘open’ state (apo structure) to a ‘closed’ state (holo structure). Many studies have utilized collections of apo–holo structure pairs to investigate the conformational transitions and critical residues. However, the collection process is usually complicated, varies from study to study and produces a small-scale data set. AH-DB is designed to provide an easy and unified way to prepare such data, which is generated by identifying/mapping molecules in different Protein Data Bank (PDB) entries. Conformational transitions are identified based on a refined alignment scheme to overcome the challenge that many structures in the PDB database are only protein fragments and not complete proteins. There are 746 314 apo–holo pairs in AH-DB, which is about 30 times those in the second largest collection of similar data. AH-DB provides sophisticated interfaces for searching apo–holo structure pairs and exploring conformational transitions from apo structures to the corresponding holo structures.