PSLDoc: Protein subcellular localization prediction based on gapped-dipeptides and probabilistic latent semantic analysis

Prediction of protein subcellular localization (PSL) is important for genome annotation, protein function prediction, and drug discovery. Many computational approaches for PSL prediction based on protein sequences have been proposed in recent years for Gram-negative bacteria. We present PSLDoc, a method based on gapped-dipeptides and probabilistic latent semantic analysis (PLSA) to solve this problem. A protein is considered as a term string composed by gapped-dipeptides, which are defined as any two residues separated by one or more positions. The weighting scheme of gapped-dipeptides is calculated according to a position specific score matrix, which includes sequence evolutionary information. Then, PLSA is applied for feature reduction, and reduced vectors are input to five one-versus-rest support vector machine classifiers. The localization site with the highest probability is assigned as the final prediction. It has been reported that there is a strong correlation between sequence homology and subcellular localization (Nair and Rost, Protein Sci 2002;11:2836-2847; Yu et al., Proteins 2006;64:643-651). To properly evaluate the performance of PSLDoc, a target protein can be classified into low- or high-homology data sets. PSLDoc's overall accuracy of low- and high-homology data sets reaches 86.84% and 98.21%, respectively, and it compares favorably with that of CELLO II (Yu et al., Proteins 2006;64:643-651). In addition, we set a confidence threshold to achieve a high precision at specified levels of recall rates. When the confidence threshold is set at 0.7, PSLDoc achieves 97.89% in precision which is considerably better than that of PSORTb v.2.0 (Gardy et al., Bioinformatics 2005;21:617-623). Our approach demonstrates that the specific feature representation for proteins can be successfully applied to the prediction of protein subcellular localization and improves prediction accuracy. Besides, because of the generality of the representation, our method can be extended to eukaryotic proteomes in the future. The web server of PSLDoc is publicly available at http://bio-cluster.iis.sinica.edu.tw/~ bioapp/PSLDoc/.     

Remote homology detection based on oligomer distances

MOTIVATION: Remote homology detection is among the most intensively researched problems in bioinformatics. Currently discriminative approaches, especially kernel-based methods, provide the most accurate results. However, kernel methods also show several drawbacks: in many cases prediction of new sequences is computationally expensive, often kernels lack an interpretable model for analysis of characteristic sequence features, and finally most approaches make use of so-called hyperparameters which complicate the application of methods across different datasets. RESULTS: We introduce a feature vector representation for protein sequences based on distances between short oligomers. The corresponding feature space arises from distance histograms for any possible pair of K-mers. Our distance-based approach shows important advantages in terms of computational speed while on common test data the prediction performance is highly competitive with state-of-the-art methods for protein remote homology detection. Furthermore the learnt model can easily be analyzed in terms of discriminative features and in contrast to other methods our representation does not require any tuning of kernel hyperparameters. AVAILABILITY: Normalized kernel matrices for the experimental setup can be downloaded at www.gobics.de/thomas. Matlab code for computing the kernel matrices is available upon request. CONTACT: thomas@gobics.de, peter@gobics.de.     

The prediction of protein structural class using averaged chemical shifts

Knowledge of protein structural class can provide important information about its folding patterns. Many approaches have been developed for the prediction of protein structural classes. However, the information used by these approaches is primarily based on amino acid sequences. In this study, a novel method is presented to predict protein structural classes by use of chemical shift (CS) information derived from nuclear magnetic resonance spectra. Firstly, 399 non-homologue (about 15% identity) proteins were constructed to investigate the distribution of averaged CS values of six nuclei ((13)CO, (13)Cα, (13)Cβ, (1)HN, (1)Hα and (15)N) in three protein structural classes. Subsequently, support vector machine was proposed to predict three protein structural classes by using averaged CS information of six nuclei. Overall accuracy of jackknife cross-validation achieves 87.0%. Finally, the feature selection technique is applied to exclude redundant information and find out an optimized feature set. Results show that the overall accuracy increased to 88.0% by using the averaged CSs of (13)CO, (1)Hα and (15)N. The proposed approach outperformed other state-of-the-art methods in terms of predictive accuracy in particular for low-similarity protein data. We expect that our proposed approach will be an excellent alternative to traditional methods for protein structural class prediction.     

Classifier ensembles for protein structural class prediction with varying homology

Structural class characterizes the overall folding type of a protein or its domain. A number of computational methods have been proposed to predict structural class based on primary sequences; however, the accuracy of these methods is strongly affected by sequence homology. This paper proposes, an ensemble classification method and a compact feature-based sequence representation. This method improves prediction accuracy for the four main structural classes compared to competing methods, and provides highly accurate predictions for sequences of widely varying homologies. The experimental evaluation of the proposed method shows superior results across sequences that are characterized by entire homology spectrum, ranging from 25% to 90% homology. The error rates were reduced by over 20% when compared with using individual prediction methods and most commonly used composition vector representation of protein sequences. Comparisons with competing methods on three large benchmark datasets consistently show the superiority of the proposed method.     

Prediction of protein structural classes by Chou's pseudo amino acid composition: approached using continuous wavelet transform and principal component analysis

A prior knowledge of protein structural classes can provide useful information about its overall structure, so it is very important for quick and accurate determination of protein structural class with computation method in protein science. One of the key for computation method is accurate protein sample representation. Here, based on the concept of Chou’s pseudo-amino acid composition (AAC, Chou, Proteins: structure, function, and genetics, 43:246–255, 2001), a novel method of feature extraction that combined continuous wavelet transform (CWT) with principal component analysis (PCA) was introduced for the prediction of protein structural classes. Firstly, the digital signal was obtained by mapping each amino acid according to various physicochemical properties. Secondly, CWT was utilized to extract new feature vector based on wavelet power spectrum (WPS), which contains more abundant information of sequence order in frequency domain and time domain, and PCA was then used to reorganize the feature vector to decrease information redundancy and computational complexity. Finally, a pseudo-amino acid composition feature vector was further formed to represent primary sequence by coupling AAC vector with a set of new feature vector of WPS in an orthogonal space by PCA. As a showcase, the rigorous jackknife cross-validation test was performed on the working datasets. The results indicated that prediction quality has been improved, and the current approach of protein representation may serve as a useful complementary vehicle in classifying other attributes of proteins, such as enzyme family class, subcellular localization, membrane protein types and protein secondary structure, etc.     

Prediction of protein structural classes by recurrence quantification analysis based on chaos game representation

In this paper, we intend to predict protein structural classes (α, β, α+β, or α/β) for low-homology data sets. Two data sets were used widely, 1189 (containing 1092 proteins) and 25PDB (containing 1673 proteins) with sequence homology being 40% and 25%, respectively. We propose to decompose the chaos game representation of proteins into two kinds of time series. Then, a novel and powerful nonlinear analysis technique, recurrence quantification analysis (RQA), is applied to analyze these time series. For a given protein sequence, a total of 16 characteristic parameters can be calculated with RQA, which are treated as feature representation of protein sequences. Based on such feature representation, the structural class for each protein is predicted with Fisher's linear discriminant algorithm. The jackknife test is used to test and compare our method with other existing methods. The overall accuracies with step-by-step procedure are 65.8% and 64.2% for 1189 and 25PDB data sets, respectively. With one-against-others procedure used widely, we compare our method with five other existing methods. Especially, the overall accuracies of our method are 6.3% and 4.1% higher for the two data sets, respectively. Furthermore, only 16 parameters are used in our method, which is less than that used by other methods. This suggests that the current method may play a complementary role to the existing methods and is promising to perform the prediction of protein structural classes.     

The use of functional domains to improve transmembrane protein topology prediction

Transmembrane proteins affect vital cellular functions and pathogenesis, and are a focus of drug design. It is difficult to obtain diffraction quality crystals to study transmembrane protein structure. Computational tools for transmembrane protein topology prediction fill in the gap between the abundance of transmembrane proteins and the scarcity of known membrane protein structures. Their prediction accuracy is still inadequate: TMHMM, the current state-of-the-art method, has less than 52% accuracy in topology prediction on one set of transmembrane proteins of known topology. Based on the observation that there are functional domains that occur preferentially internal or external to the membrane, we have extended the model of TMHMM to incorporate functional domains, using a probabilistic approach originally developed for computational gene finding. Our extension is better than TMHMM in predicting the topology of transmembrane proteins. As prediction of functional domain improves, our system's prediction accuracy will likely improve as well.     

Carnosic acid and carnosol inhibit adipocyte differentiation in mouse 3T3-L1 cells through induction of phase2 enzymes and activation of glutathione metabolism

In the previous studies, we reported that carnosic acid (CA) and carnosol (CS) originating from rosemary protected cortical neurons by activating the Keap1/Nrf2 pathway, which activation was initiated by S-alkylation of the critical cysteine thiol of the Keap1 protein by the “electrophilic” quinone-type of CA or CS. Here, we found that CA and CS inhibited the in vitro differentiation of mouse preadipocytes, 3T3-L1 cells, into adipocytes. In contrast, other physiologically-active and rosemary-originated compounds were completely negative. These actions seemed to be mediated by activation of the antioxidant-response element (ARE) and induction of phase2 enzymes. This estimation is justified by our present findings that only CA and CS among rosemary-originated compounds significantly activated the ARE and induced the phase2 enzymes. Next, we performed cDNA microarray analysis in order to identify the gene(s) responsible for these biological actions and found that phase2 enzymes (Gsta2, Gclc, Abcc4, and Abcc1), all of which are involved in the metabolism of glutathione (GSH), constituted 4 of the top 5 CA-induced genes. Furthermore, CA and CS, but not the other compounds tested, significantly increased the intracellular level of total GSH. Thus, we propose that the stimulation of GSH metabolism may be a critical step for the inhibition of adipocyte differentiation in 3T3-L1 cells and suggest that pro-electrophilic compounds such as CA and CS may be potential drugs against obesity-related diseases.     

Explainable deep transfer learning model for disease risk prediction using high-dimensional genomic data

Building an accurate disease risk prediction model is an essential step in the modern quest for precision medicine. While high-dimensional genomic data provides valuable data resources for the investigations of disease risk, their huge amount of noise and complex relationships between predictors and outcomes have brought tremendous analytical challenges. Deep learning model is the state-of-the-art methods for many prediction tasks, and it is a promising framework for the analysis of genomic data. However, deep learning models generally suffer from the curse of dimensionality and the lack of biological interpretability, both of which have greatly limited their applications. In this work, we have developed a deep neural network (DNN) based prediction modeling framework. We first proposed a group-wise feature importance score for feature selection, where genes harboring genetic variants with both linear and non-linear effects are efficiently detected. We then designed an explainable transfer-learning based DNN method, which can directly incorporate information from feature selection and accurately capture complex predictive effects. The proposed DNN-framework is biologically interpretable, as it is built based on the selected predictive genes. It is also computationally efficient and can be applied to genome-wide data. Through extensive simulations and real data analyses, we have demonstrated that our proposed method can not only efficiently detect predictive features, but also accurately predict disease risk, as compared to many existing methods.     

Identification and functional characterization of mRNAs that exhibit stop codon readthrough in Arabidopsis thaliana

Stop codon readthrough (SCR) is the process of continuation of translation beyond the stop codon, generating protein isoforms with C-terminal extensions. SCR has been observed in viruses, fungi, and multicellular organisms, including mammals. However, SCR is largely unexplored in plants. In this study, we have analyzed ribosome profiling datasets to identify mRNAs that exhibit SCR in Arabidopsis thaliana. Analyses of the ribosome density, ribosome coverage, and three-nucleotide periodicity of the ribosome profiling reads in the mRNA region downstream of the stop codon provided strong evidence for SCR in mRNAs of 144 genes. We show that SCR generated putative evolutionarily conserved nuclear localization signals, transmembrane helices, and intrinsically disordered regions in the C-terminal extensions of several of these proteins. Furthermore, gene ontology functional enrichment analysis revealed that these 144 genes belong to three major functional groups—translation, photosynthesis, and abiotic stress tolerance. Using a luminescence-based readthrough assay, we experimentally demonstrated SCR in representative mRNAs belonging to each of these functional classes. Finally, using microscopy, we show that the SCR product of one gene that contains a nuclear localization signal at the C-terminal extension, CURT1B, localizes to the nucleus as predicted. Based on these observations, we propose that SCR plays an important role in plant physiology by regulating protein localization and function.     

Comparative analysis of an experimental subcellular protein localization assay and in silico prediction methods

The subcellular localization of a protein can provide important information about its function within the cell. As eukaryotic cells and particularly mammalian cells are characterized by a high degree of compartmentalization, most protein activities can be assigned to particular cellular compartments. The categorization of proteins by their subcellular localization is therefore one of the essential goals of the functional annotation of the human genome. We previously performed a subcellular localization screen of 52 proteins encoded on human chromosome 21. In the current study, we compared the experimental localization data to the in silico results generated by nine leading software packages with different prediction resolutions. The comparison revealed striking differences between the programs in the accuracy of their subcellular protein localization predictions. Our results strongly suggest that the recently developed predictors utilizing multiple prediction methods tend to provide significantly better performance over purely sequence-based or homology-based predictions.     

Accurate Prediction of Protein Catalytic Residues by Side Chain Orientation and Residue Contact Density

Prediction of protein catalytic residues provides useful information for the studies of protein functions. Most of the existing methods combine both structure and sequence information but heavily rely on sequence conservation from multiple sequence alignments. The contribution of structure information is usually less than that of sequence conservation in existing methods. We found a novel structure feature, residue side chain orientation, which is the first structure-based feature that achieves prediction results comparable to that of evolutionary sequence conservation. We developed a structure-based method, Enzyme Catalytic residue SIde-chain Arrangement (EXIA), which is based on residue side chain orientations and backbone flexibility of protein structure. The prediction that uses EXIA outperforms existing structure-based features. The prediction quality of combing EXIA and sequence conservation exceeds that of the state-of-the-art prediction methods. EXIA is designed to predict catalytic residues from single protein structure without needing sequence or structure alignments. It provides invaluable information when there is no sufficient or reliable homology information for target protein. We found that catalytic residues have very special side chain orientation and designed the EXIA method based on the newly discovered feature. It was also found that EXIA performs well for a dataset of enzymes without any bounded ligand in their crystallographic structures.     

Predicting subcellular localization of proteins for Gram-negative bacteria by support vector machines based on n-peptide compositions

Gram-negative bacteria have five major subcellular localization sites: the cytoplasm, the periplasm, the inner membrane, the outer membrane, and the extracellular space. The subcellular location of a protein can provide valuable information about its function. With the rapid increase of sequenced genomic data, the need for an automated and accurate tool to predict subcellular localization becomes increasingly important. We present an approach to predict subcellular localization for Gram-negative bacteria. This method uses the support vector machines trained by multiple feature vectors based on n-peptide compositions. For a standard data set comprising 1443 proteins, the overall prediction accuracy reaches 89%, which, to the best of our knowledge, is the highest prediction rate ever reported. Our prediction is 14% higher than that of the recently developed multimodular PSORT-B. Because of its simplicity, this approach can be easily extended to other organisms and should be a useful tool for the high-throughput and large-scale analysis of proteomic and genomic data.     

Dimeric rous sarcoma virus capsid protein structure relevant to immature Gag assembly

The structure of the N-terminal domain (NTD) of Rous sarcoma virus (RSV) capsid protein (CA), with an upstream 25 amino acid residue extension corresponding to the C-terminal portion of the Gag p10 protein, has been determined by X-ray crystallography. Purified Gag proteins of retroviruses can assemble in vitro into virus-like particles closely resembling in vivo-assembled immature virus particles, but without a membrane. When the 25 amino acid residues upstream of CA are deleted, Gag assembles into tubular particles. The same phenotype is observed in vivo. Thus, these residues act as a “shape determinant” promoting spherical assembly, when they are present, or tubular assembly, when they are absent. We show that, unlike the NTD on its own, the extended NTD protein has no β-hairpin loop at the N terminus of CA and that the molecule forms a dimer in which the amino-terminal extension forms the interface between monomers. Since dimerization of Gag has been inferred to be a critical step in assembly of spherical, immature Gag particles, the dimer interface may represent a structural feature that is essential in retrovirus assembly.     

Cloning,characterization and expression of a cDNA encoding a granulin-like polypeptide in Ciona savignyi

Previous study in our laboratory confirmed that a novel polypeptide, CS5931 derived from Ciona savignyi possesses potent antitumor activity. In the present study, the full length cDNA of CS5931 precursor, termed Cs-pgrn-1 was cloned. The complete cDNA sequence of this gene consists of 685 bp containing an open reading frame (ORF) of 522 bp (173 amino acid residues). In silico analysis revealed that the polypeptide consists of two identical domains, similar with granulin (GRN) found in other species, and each of the domain encodes a polypeptide identical with CS5931. Phylogenetic analysis confirmed that CS5931 shares high homology with Ciona intestinalis GRN and is conserved during evolution. The polypeptide also shows high similarity with human GRN A, B, and C. Prediction of 3D protein structure revealed the 3D structure of CS5931 is very similar with human GRN A. The CS5931 was expressed using a prokaryotic expression system and the purified polypeptide inhibited the growth of several tumor cell lines in vitro via apoptotic pathway. Our study revealed that CS5931 has the potential to be developed as a novel antitumor agent.     

Using Amino Acid Physicochemical Distance Transformation for Fast Protein Remote Homology Detection

Protein remote homology detection is one of the most important problems in bioinformatics. Discriminative methods such as support vector machines (SVM) have shown superior performance. However, the performance of SVM-based methods depends on the vector representations of the protein sequences. Prior works have demonstrated that sequence-order effects are relevant for discrimination, but little work has explored how to incorporate the sequence-order information along with the amino acid physicochemical properties into the prediction. In order to incorporate the sequence-order effects into the protein remote homology detection, the physicochemical distance transformation (PDT) method is proposed. Each protein sequence is converted into a series of numbers by using the physicochemical property scores in the amino acid index (AAIndex), and then the sequence is converted into a fixed length vector by PDT. The sequence-order information can be efficiently included into the feature vector with little computational cost by this approach. Finally, the feature vectors are input into a support vector machine classifier to detect the protein remote homologies. Our experiments on a well-known benchmark show the proposed method SVM-PDT achieves superior or comparable performance with current state-of-the-art methods and its computational cost is considerably superior to those of other methods. When the evolutionary information extracted from the frequency profiles is combined with the PDT method, the profile-based PDT approach can improve the performance by 3.4% and 11.4% in terms of ROC score and ROC50 score respectively. The local sequence-order information of the protein can be efficiently captured by the proposed PDT and the physicochemical properties extracted from the amino acid index are incorporated into the prediction. The physicochemical distance transformation provides a general framework, which would be a valuable tool for protein-level study.     

Predicting protein function by multi-label correlated semi-supervised learning

Assigning biological functions to uncharacterized proteins is a fundamental problem in the postgenomic era. The increasing availability of large amounts of data on protein-protein interactions (PPIs) has led to the emergence of a considerable number of computational methods for determining protein function in the context of a network. These algorithms, however, treat each functional class in isolation and thereby often suffer from the difficulty of the scarcity of labeled data. In reality, different functional classes are naturally dependent on one another. We propose a new algorithm, Multi-label Correlated Semi-supervised Learning (MCSL), to incorporate the intrinsic correlations among functional classes into protein function prediction by leveraging the relationships provided by the PPI network and the functional class network. The guiding intuition is that the classification function should be sufficiently smooth on subgraphs where the respective topologies of these two networks are a good match. We encode this intuition as regularized learning with intraclass and interclass consistency, which can be understood as an extension of the graph-based learning with local and global consistency (LGC) method. Cross validation on the yeast proteome illustrates that MCSL consistently outperforms several state-of-the-art methods. Most notably, it effectively overcomes the problem associated with scarcity of label data. The supplementary files are freely available at http://sites.google.com/site/csaijiang/MCSL.     

Protein location prediction using atomic composition and global features of the amino acid sequence

Subcellular location of protein is constructive information in determining its function, screening for drug candidates, vaccine design, annotation of gene products and in selecting relevant proteins for further studies. Computational prediction of subcellular localization deals with predicting the location of a protein from its amino acid sequence. For a computational localization prediction method to be more accurate, it should exploit all possible relevant biological features that contribute to the subcellular localization. In this work, we extracted the biological features from the full length protein sequence to incorporate more biological information. A new biological feature, distribution of atomic composition is effectively used with, multiple physiochemical properties, amino acid composition, three part amino acid composition, and sequence similarity for predicting the subcellular location of the protein. Support Vector Machines are designed for four modules and prediction is made by a weighted voting system. Our system makes prediction with an accuracy of 100, 82.47, 88.81 for self-consistency test, jackknife test and independent data test respectively. Our results provide evidence that the prediction based on the biological features derived from the full length amino acid sequence gives better accuracy than those derived from N-terminal alone. Considering the features as a distribution within the entire sequence will bring out underlying property distribution to a greater detail to enhance the prediction accuracy.     

Widely predicting specific protein functions based on protein-protein interaction data and gene expression profile

GESTs (gene expression similarity and taxonomy similarity), a gene functional prediction approach previously proposed by us, is based on gene expression similarity and concept similarity of functional classes defined in Gene Ontology (GO). In this paper, we extend this method to protein-protein interaction data by introducing several methods to filter the neighbors in protein interaction networks for a protein of unknown function(s). Unlike other conventional methods, the proposed approach automatically selects the most appropriate functional classes as specific as possible during the learning process, and calls on genes annotated to nearby classes to support the predictions to some small-sized specific classes in GO. Based on the yeast protein-protein interaction information from MIPS and a dataset of gene expression profiles, we assess the performances of our approach for predicting protein functions to “biology process” by three measures particularly designed for functional classes organized in GO. Results show that our method is powerful for widely predicting gene functions with very specific functional terms. Based on the GO database published in December 2004, we predict some proteins whose functions were unknown at that time, and some of the predictions have been confirmed by the new SGD annotation data published in April, 2006.