For Protein secondary structure

Beta-turns play an important role in protein structures not only because of their sheer abundance, which is estimated to be approximately 25% of all protein residues, but also because of their significance in high-order structures of proteins. In this study, we introduce a new method of beta-turn prediction that uses a two-stage classification scheme and an integrated framework for input features. Ten-fold cross validation based on a benchmark dataset of 426 non-homologue protein chains is used to evaluate our method’s performance. The experimental results demonstrate that it achieves substantial improvements over the current best method.


(1) A knowledge-based hybrid method for protein secondary structure prediction based on local prediction confidence

*Motivation:*

In our previous approach, we proposed a hybrid method for protein secondary structure prediction, called HYPROSP, which combined our proposed knowledge-based prediction algorithm PROSP and PSIPRED. The knowledge base constructed for PROSP contains small peptides together with their secondary structural information. The hybrid strategy of HYPROSP uses a global quantitative measure, *match rate* to determine which of PROSP and PSIPRED to be used for prediction of a target protein. HYPROSP made slight improvement of Q3 over PSIPRED because PROSP predicted well for proteins with match rate above 80%. As the portion of proteins with match rate above 80% is quite small and the performance of PSIPRED also improves, the advantage of HYPROSP is diluted. To overcome this limitation and further improve the hybrid prediction method, we present in this paper a new hybrid strategy HYPROSP II which is based on a new quantitative measure called *local match rate*.

*Results:*

Local match rate indicates the amount of structural information that each amino acid can extract from the knowledge base. With the local match rate, we are able to define a confidence level of the PROSP prediction results for each amino acid. Our new hybrid approach, HYPROSP II, is proposed as follows: for each amino acid in a target protein, we combine the prediction results of PROSP and PSIPRED using a hybrid function defined on their respective confidence levels. Two datasets in nrDSSP and EVA are used to perform a tenfold cross validation. The average Q3 of HYPROSP II is 81.8% and 80.7% on nrDSSP and EVA datasets, respectively, which is 2.0% and 1.1% better than that of PSIPRED. For local structures with match rate higher than 80%, the average Q3 improvement is 4.4% on the nrDSSP dataset. Using local match rate improves the accuracy better than global match rate. There has been a long history of attempts to improve secondary structure prediction. We believe HYPROSP II has greatly utilized the power of peptide knowledge base and raised the prediction accuracy to a new high. The method we developed in this paper could have a profound effect on the general use of knowledge base techniques for various prediction algorithms.

(2) HYPROSP: a hybrid protein secondary structure prediction algorithm—a knowledge-based approach

We develop a knowledge-based approach (called PROSP) for protein secondary structure prediction. The knowledge base contains small peptide fragments together with their secondary structural information. A quantitative measure M, called match rate, is defined to measure the amount of structural information that a target protein can extract from the knowledge base. Our experimental results show that proteins with a higher match rate will likely be predicted more accurately based on PROSP. That is, there is roughly a monotone correlation between the prediction accuracy and the amount of structure matching with the knowledge base. To fully utilize the strength of our knowledge base, a hybrid prediction method is proposed as follows: if the match rate of a target protein is at least 80%, we use the extracted information to make the prediction; otherwise, we adopt a popular machine-learning approach. This comprises our hybrid protein structure prediction (HYPROSP) approach. We use the DSSP and EVA data as our datasets and PSIPRED as our underlying machine-learning algorithm. For target proteins with match rate at least 80%, the average Q3 of PROSP is 3.96 and 7.2 better than that of PSIPRED on DSSP and EVA data, respectively.

For Protein Local Structure


(3) A Knowledge-based Approach to Protein Local Structure Prediction

Local structure prediction is an important step towards protein tertiary structure prediction. We build up a sequence-structure knowledge base which stores information of local structures and amino acids. Then a hybrid method based on our knowledge base and neural network is proposed for predicting backbone local structure of proteins. Our method yields over 60% accuracy, which is significantly higher than others. Results also demonstrates that our approach is applicable to different local structure libraries. Related paper of our work appear in Journal of Bioinformatics and Computational Biology 2006.


For Protein beta-turn


(4) A two-stage classifier for protein beta-turn prediction using support vector machines

Beta-turns play an important role in protein structures not only because of their sheer abundance, which is estimated to be approximately 25% of all protein residues, but also because of their significance in high-order structures of proteins. In this study, we introduce a new method of beta-turn prediction that uses a two-stage classification scheme and an integrated framework for input features. Ten-fold cross validation based on a benchmark dataset of 426 non-homologue protein chains is used to evaluate our method’s performance. The experimental results demonstrate that it achieves substantial improvements over the current best method.

Protein subcellular localization prediction The study of protein subcellular localization (PSL) is important for elucidating protein functions involved in various cellular processes. We proposed two PSL prediction methods, PSL101 and PSLDoc, which were published in BMC Bioinformatics 2007 and Proteins: Structure, Function, and Bioinformatics 2008, respectively.

PSL101 combines a structural homology approach and a support vector machine model, in which biological features derived from bacterial translocation pathways are incorporated. PSLDoc uses a probabilistic latent semantic analysis on gapped-dipeptides of various distances, where evolutionary information from position specific scoring matrix is utilized. Our methods achieve 79%~94% in overall accuracy for several prokaryotic and eukaryotic species, and compared favorably to the respective state-of-the-art results by 7.4% when those sequences of low homology to the training set. Moreover, the proposed biological features and gapped-dipeptide signatures are interpretable and can be applied in advanced studies and experimental designs.

(1) PSLDoc: Protein subcellular localization prediction based on gapped-dipeptides and probabilistic latent semantic analysis
(2) Protein subcellular localization prediction based on compartment-specific features and structure conservation
(3) Protein subcellular localization prediction based on compartment-specific biological features

Factor Xa. (Ei-Wen Yang, Ching-Tai Chen)

Regulatory proteases modulate biological systems through catalyzing hydrolysis reactions on designated peptide bonds, which in turn tip proteomic balance, resulting in rapid and substantial change of the systems. The regulatory power of the proteases is known to originate from the corresponding substrate specificities of the proteases. However, for large part of the protease families, the specificity spectrum and the in vivo substrates have been poorly comprehended. A viable approach for elucidating the proteomic role of a regulatory protease is to construct computational systems capable of modeling the rules governing the complete substrate specificities of the protease to scan for potential in vivo substrates in the proteome. The predicted substrate sites could provide important clues for the protease's regulatory network, on which verifying experiments could then be focused. We have constructed computational models for scanning potential substrate sites in nature protein sequences by integrating multi-level substrate phage display experiments with bootstrap aggregation machine learning algorithms. Factor Xa, a key regulatory serine protease of the blood coagulation system, was used as a model protease to demonstrate that the systematically coupled experimental and computational procedures together were able to produce computational systems capable of scanning for substrate sites of significant biological relevance in nature protein sequences. The protocol for experimentally sampling and computationally learning on the rules governing the substrate specificity can be generalized to any protease of interest for which the active form is available for the in vitro experiments.

Current fully-sequenced genomes encode 20%-30% integral membrane proteins (IMPs) and they are critical in many biological activities and serve as important drug targets. However, the number of their high-resolution structures is limited due to experimental difficulties. Therefore, in silico prediction and modeling play a key role in elucidating the structure genomics of IMPs. As the first step in understanding IMP structures, we have developed SVMtop, a hierarchical classification method using support vector machines (SVMs) to predict the topology models of IMPs. The prediction framework integrates selected biological features and a novel topology scoring function based on membrane protein folding. Standard benchmarks show that SVMtop ranks highly among the methods compared and achieves approximately 70% accuracy for correctly predicting both the number of TM helix and topology, and less than 1% of false positive rate for identifying soluble proteins. Building upon this, we are currently investigating interactions between TM helices which are a major force in membrane protein assembly. We are developing computational algorithms to predict helix-helix interactions which may in turn, aid in structure prediction of membrane proteins.

(1) Enhanced membrane protein topology prediction using a hierarchical classification method and a new scoring function
(2) On the accuracy of transmembrane helix prediction methods using an updated benchmark
(3) Transmembrane Helix and Topology Prediction Using Hierarchical SVM Classifiers and an Alternating Geometric Scoring Function
(4) Predicting helix-helix interactions from residue contacts in membrane proteins

NMR data from different experiments often contain errors; thus, automated backbone resonance assignment is a very challenging issue. We develop an iterative relaxation algorithm, called RIBRA, for NMR protein backbone assignment. RIBRA applies nearest neighbor and weighted maximum independent set algorithms to solve the problem. We test RIBRA on two real NMR datasets: hbSBD and hbLBD, and perfect BMRB data (with 902 proteins) and four synthetic BMRB data which simulate four kinds of errors. The accuracy of RIBRA on hb-SBD and hbLBD are 91.4% and 83.6%, respectively. The average accuracy of RIBRA on perfect BMRB datasets is 98.28%, and 98.28%, 95.61%, 98.16% and 96.28% on four kinds of synthetic datasets, respectively.

Besides, we also present a method called GANA that uses a genetic algorithm to automatically perform backbone resonance assignment with a high degree of precision and recall. GANA takes spin systems as input data and almost all spin systems can be mapped correctly onto a target protein, even if the data are noisy. The average recall rates of GANA on BMRB and the four simulated test cases are 99.26, 99.19, 98.85, 98.87 and 97.78%, respectively. The precision and recall rates of GANA on hbSBD are 95.12 and 92.86%, respectively, and those of hbLBD are 100 and 97.40%, respectively.

(1) RIBRA-an Error-Tolerant Algorithm for the NMR Backbone Assignment Problem
(2) GANA – A Genetic Algorithm for NMR Backbone Resonance Assignment
(3) An Iterative Relaxation Technique for the NMR Backbone Assignment Problem

RNA-protein binding prediction

RNA-protein interaction plays an essential role in several biological processes, such as protein synthesis, gene expression, post-transcriptional regulation, and antiviral drug discovery. Identification of RNA-binding sites in proteins can provide valuable insights for biologists. However, experimental determination RNA-protein interaction remains time-consuming and labor-intensive. Thus, computational approaches for the prediction of RNA-binding sites from protein sequences have become highly desirable. In this paper, we propose a method, RNAProB, to predict RNA-binding sites based on support vector machines and a new encoding scheme for smoothed position-specific scoring matrix. Evaluated by five-fold cross-validation, our method achieves Matthew’s correlation coefficient (MCC) values of 0.68, 0.58, and 0.42 compared to 0.45, 0.35, and 0.32 by the state-of-the-art systems for three benchmark data sets, respectively. Moreover, to avoid data overfitting, we use a three-way data split procedure to estimate our predictive performance, and our approach obtains MCC values of 0.67, 0.56, and 0.40, respectively. In conclusion, our method significantly improves the predictive performance of RNA-binding site prediction. The proposed encoding scheme for smoothed PSSM can be used in other research problems, such as DNA-protein interaction, protein-protein interaction, and prediction of post-translational modification, etc.