Predicting Metabolic Pathways of Small Molecules and Enzymes Based on Interaction Information of Chemicals and Proteins

Metabolic pathway analysis, one of the most important fields in biochemistry, is pivotal to understanding the maintenance and modulation of the functions of an organism. Good comprehension of metabolic pathways is critical to understanding the mechanisms of some fundamental biological processes. Given a small molecule or an enzyme, how may one identify the metabolic pathways in which it may participate? Answering such a question is a first important step in understanding a metabolic pathway system. By utilizing the information provided by chemical-chemical interactions, chemical-protein interactions, and protein-protein interactions, a novel method was proposed by which to allocate small molecules and enzymes to 11 major classes of metabolic pathways. A benchmark dataset consisting of 3,348 small molecules and 654 enzymes of yeast was constructed to test the method. It was observed that the first order prediction accuracy evaluated by the jackknife test was 79.56% in identifying the small molecules and enzymes in a benchmark dataset. Our method may become a useful vehicle in predicting the metabolic pathways of small molecules and enzymes, providing a basis for some further analysis of the pathway systems.     

Prediction of the functional roles of small molecules in lipid metabolism based on ensemble learning

As many diseases like high cholesterol are referred to lipid metabolism, studying the lipid metabolic pathway has a positive effect on finding the knowledge about interactions between different elements within high complex living systems. Here, we employed a typical ensemble learning method, Bagging learner, to study and predict the possible sub lipid metabolic pathway of small molecules based on physical and chemical features of the compounds. As a result, jackknife cross validation test and independent set test on the model reached 89.85% and 91.46%, respectively. Therefore, our predictor may be used for finding the new compounds which participate in lipid metabolic procedures.     

Protein binding site prediction using an empirical scoring function

Most biological processes are mediated by interactions between proteins and their interacting partners including proteins, nucleic acids and small molecules. This work establishes a method called PINUP for binding site prediction of monomeric proteins. With only two weight parameters to optimize, PINUP produces not only 42.2% coverage of actual interfaces (percentage of correctly predicted interface residues in actual interface residues) but also 44.5% accuracy in predicted interfaces (percentage of correctly predicted interface residues in the predicted interface residues) in a cross validation using a 57-protein dataset. By comparison, the expected accuracy via random prediction (percentage of actual interface residues in surface residues) is only 15%. The binding sites of the 57-protein set are found to be easier to predict than that of an independent test set of 68 proteins. The average coverage and accuracy for this independent test set are 30.5 and 29.4%, respectively. The significant gain of PINUP over expected random prediction is attributed to (i) effective residue-energy score and accessible-surface-area-dependent interface-propensity, (ii) isolation of functional constraints contained in the conservation score from the structural constraints through the combination of residue-energy score (for structural constraints) and conservation score and (iii) a consensus region built on top-ranked initial patches.     

Analysis of protein features and machine learning algorithms for prediction of druggable proteins

Background: Computational tools have been widely used in drug discovery process since they reduce the time and cost. Prediction of whether a protein is druggable is fundamental and crucial for drug research pipeline. Sequence based protein function prediction plays vital roles in many research areas. Training data, protein features selection and machine learning algorithms are three indispensable elements that drive the successfulness of the models. Methods: In this study, we tested the performance of different combinations of protein features and machine learning algorithms, based on FDA-approved small molecules’ targets, in druggable proteins prediction. We also enlarged the dataset to include the targets of small molecules that were in experiment or clinical investigation. Results: We found that although the 146-d vector used by Li et al. with neuron network achieved the best training accuracy of 91.10%, overlapped 3-gram word2vec with logistic regression achieved best prediction accuracy on independent test set (89.55%) and on newly approved-targets. Enlarged dataset with targets of small molecules in experiment and clinical investigation were trained. Unfortunately, the best training accuracy was only 75.48%. In addition, we applied our models to predict potential targets for references in future study. Conclusions: Our study indicates the potential ability of word2vec in the prediction of druggable protein. And the training dataset of druggable protein should not be extended to targets that are lack of verification. The target prediction package could be found on https://github.com/pkumdl/target_prediction.     

Prediction of the β-Hairpins in Proteins Using Support Vector Machine

By using of the composite vector with increment of diversity and scoring function to express the information of sequence, a support vector machine (SVM) algorithm for predicting β-hairpin motifs is proposed. The prediction is done on a dataset of 3,088 non homologous proteins containing 6,027 β-hairpins. The overall accuracy of prediction and Matthew’s correlation coefficient are 79.9% and 0.59 for the independent testing dataset. In addition, a higher accuracy of 83.3% and Matthew’s correlation coefficient of 0.67 in the independent testing dataset are obtained on a dataset previously used by Kumar et al. (Nuclic Acid Res 33:154–159). The performance of the method is also evaluated by predicting the β-hairpins of in the CASP6 proteins, and the better results are obtained. Moreover, this method is used to predict four kinds of supersecondary structures. The overall accuracy of prediction is 64.5% for the independent testing dataset.     

Is it better to combine predictions?

We have compared the accuracy of the individual protein secondary structure prediction methods: PHD, DSC, NNSSP and Predator against the accuracy obtained by combing the predictions of the methods. A range of ways of combing predictions were tested: voting, biased voting, linear discrimination, neural networks and decision trees. The combined methods that involve 'learning' (the non-voting methods) were trained using a set of 496 non-homologous domains; this dataset was biased as some of the secondary structure prediction methods had used them for training. We used two independent test sets to compare predictions: the first consisted of 17 non-homologous domains from CASP3 (Third Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction); the second set consisted of 405 domains that were selected in the same way as the training set, and were non-homologous to each other and the training set. On both test datasets the most accurate individual method was NNSSP, then PHD, DSC and the least accurate was Predator; however, it was not possible to conclusively show a significant difference between the individual methods. Comparing the accuracy of the single methods with that obtained by combing predictions it was found that it was better to use a combination of predictions. On both test datasets it was possible to obtain a approximately 3% improvement in accuracy by combing predictions. In most cases the combined methods were statistically significantly better (at P = 0.05 on the CASP3 test set, and P = 0.01 on the EBI test set). On the CASP3 test dataset there was no significant difference in accuracy between any of the combined method of prediction: on the EBI test dataset, linear discrimination and neural networks significantly outperformed voting techniques. We conclude that it is better to combine predictions.     

基于统计差表与加权投票的高精度剪接位点预测

基于机器学习的高精度剪接位点识别是真核生物基因组注释的关键.本文采用卡方测验确定序列窗口长度,构建卡方统计差表提取位置特征,并结合碱基二联体频次表征序列;针对剪接位点正负样本高度不均衡这一情形,构建10个正负样本均衡的支持向量机分类器,进行加权投票决策,有效解决了不平衡模式分类问题. HS~3D数据集上的独立测试结果显示,供体、受体位点预测准确率分别达到93.39%、90.46%,明显高于参比方法.基于卡方统计差表的位置特征能有效表征DNA序列,在分子序列信号位点识别中具有应用前景.     

Non-canonical interactions of porphyrins in porphyrin-containing proteins

In this study we have described the non-canonical interactions between the porphyrin ring and the protein part of porphyrin-containing proteins to better understand their stabilizing role. The analysis reported in this study shows that the predominant type of non-canonical interactions at porphyrins are CH···O and CH···N interactions, with a small percentage of CH···π and non-canonical interactions involving sulfur atoms. The majority of non-canonical interactions are formed from side-chains of charged and polar amino acids, whereas backbone groups are not frequently involved. The main-chain non-canonical interactions might be slightly more linear than the side-chain interactions, and they have somewhat shorter median distances. The analysis, performed in this study, shows that about 44% of the total interactions in the dataset are involved in the formation of multiple (furcated) non-canonical interactions. The high number of porphyrin–water interactions show importance of the inclusion of solvent in protein–ligand interaction studies. Furthermore, in the present study we have observed that stabilization centers are composed predominantly from nonpolar amino acid residues. Amino acids deployed in the environment of porphyrin rings are deposited in helices and coils. The results from this study might be used for structure-based porphyrin protein prediction and as scaffolds for future porphyrin-containing protein design.     

Multi-class protein fold recognition using support vector machines and neural networks

MOTIVATION: Protein fold recognition is an important approach to structure discovery without relying on sequence similarity. We study this approach with new multi-class classification methods and examined many issues important for a practical recognition system. RESULTS: Most current discriminative methods for protein fold prediction use the one-against-others method, which has the well-known 'False Positives' problem. We investigated two new methods: the unique one-against-others and the all-against-all methods. Both improve prediction accuracy by 14-110% on a dataset containing 27 SCOP folds. We used the Support Vector Machine (SVM) and the Neural Network (NN) learning methods as base classifiers. SVMs converges fast and leads to high accuracy. When scores of multiple parameter datasets are combined, majority voting reduces noise and increases recognition accuracy. We examined many issues involved with large number of classes, including dependencies of prediction accuracy on the number of folds and on the number of representatives in a fold. Overall, recognition systems achieve 56% fold prediction accuracy on a protein test dataset, where most of the proteins have below 25% sequence identity with the proteins used in training.     

A survey of genes in Eimeria tenella merozoites by EST sequencing

A study of about 500 expressed sequence tags (ESTs), derived from a merozoite cDNA library, was initiated as an approach to generate a larger pool of gene information on Eimeria tenella. Of the ESTs, 47.7% had matches with entries in the databases, including ribosomal proteins, metabolic enzymes and proteins with other functions, of which 14.3% represented previously known E. tenella genes. Thus over 50% of the ESTs had no significant database matches. The E. tenella EST dataset contained a range of highly abundant genes comparable with that found in the EST dataset of T. gondii and may thus reflect the importance of such molecules in the biology of the apicomplexan organisms. However, comparison of the two datasets revealed very few homologies between sequences of apical organelle molecules, and provides evidence for sequence divergence between these closely-related parasites. The data presented underpin the potential value of the EST strategy for the discovery of novel genes and may allow for a more rapid increase in the knowledge and understanding of gene expression in the merozoite life cycle stage of Eimeria spp.     

Identifying GPCRs and their types with Chou's pseudo amino acid composition: an approach from multi-scale energy representation and position specific scoring matrix

G-protein coupled receptor (GPCR) is a membrane protein family, which serves as an interface between cell and the outside world. They are involved in various physiological processes and are the targets of more than 50% of the marketed drugs. The function of GPCRs can be known by conducting Biological experiments. However, the rapid increase of GPCR sequences entering into databanks, it is very time consuming and expensive to determine their function based only on experimental techniques. Hence, the computational prediction of GPCRs is very much demanding for both pharmaceutical and educational research. Feature extraction of GPCRs in the proposed research is performed using three techniques i.e. Pseudo amino acid composition, Wavelet based multi-scale energy and Evolutionary information based feature extraction by utilizing the position specific scoring matrices. For classification purpose, a majority voting based ensemble method is used; whose weights are optimized using genetic algorithm. Four classifiers are used in the ensemble i.e. Nearest Neighbor, Probabilistic Neural Network, Support Vector Machine and Grey Incidence Degree. The performance of the proposed method is assessed using Jackknife test for a number of datasets. First, the individual performances of classifiers are assessed for each dataset using Jackknife test. After that, the performance for each dataset is improved by using weighted ensemble classification. The weights of ensemble are optimized using various runs of Genetic Algorithm. We have compared our method with various other methods. The significance in performance of the proposed method depicts it to be useful for GPCRs classification.     

Het-PDB Navi.: a database for protein-small molecule interactions

The genomes of more than 100 species have been sequenced, and the biological functions of encoded proteins are now actively being researched. Protein function is based on interactions between proteins and other molecules. One approach to assuming protein function based on genomic sequence is to predict interactions between an encoded protein and other molecules. As a data source for such predictions, knowledge regarding known protein-small molecule interactions needs to be compiled. We have, therefore, surveyed interactions between proteins and other molecules in Protein Data Bank (PDB), the protein three-dimensional (3D) structure database. Among 20,685 entries in PDB (April, 2003), 4,189 types of small molecules were found to interact with proteins. Biologically relevant small molecules most often found in PDB were metal ions, such as calcium, zinc, and magnesium. Sugars and nucleotides were the next most common. These molecules are known to act as cofactors for enzymes and/or stabilizers of proteins. In each case of interactions between a protein and small molecule, we found preferred amino acid residues at the interaction sites. These preferences can be the basis for predicting protein function from genomic sequence and protein 3D structures. The data pertaining to these small molecules were collected in a database named Het-PDB Navi., which is freely available at http://daisy.nagahama-i-bio.ac.jp/golab/hetpdbnavi.html and linked to the official PDB home page.     

Looking at enzymes from the inside out: the proximity of catalytic residues to the molecular centroid can be used for detection of active sites and enzyme-ligand interfaces

Analysis of the distances of the exposed residues in 175 enzymes from the centroids of the molecules indicates that catalytic residues are very often found among the 5% of residues closest to the enzyme centroid. This property of catalytic residues is implemented in a new prediction algorithm (named EnSite) for locating the active sites of enzymes and in a new scheme for re-ranking enzyme-ligand docking solutions. EnSite examines only 5% of the molecular surface (represented by surface dots) that is closest to the centroid, identifying continuous surface segments and ranking them by their area size. EnSite ranks the correct prediction 1-4 in 97% of the cases in a dataset of 65 monomeric enzymes (rank 1 for 89% of the cases) and in 86% of the cases in a dataset of 176 monomeric and multimeric enzymes from all six top-level enzyme classifications (rank 1 in 74% of the cases). Importantly, identification of buried or flat active sites is straightforward because EnSite "looks" at the molecular surface from the inside out. Detailed examination of the results indicates that the proximity of the catalytic residues to the centroid is a property of the functional unit, defined as the assembly of domains or chains that form the active site (in most cases the functional unit corresponds to a single whole polypeptide chain). Using the functional unit in the prediction further improves the results. The new property of active sites is also used for re-evaluating enzyme-inhibitor unbound docking results. Sorting the docking solutions by the distance of the interface to the centroid of the enzyme improves remarkably the ranks of nearly correct solutions compared to ranks based on geometric-electrostatic-hydrophobic complementarity scores.     

A systematic prediction of multiple drug-target interactions from chemical, genomic, and pharmacological data

In silico prediction of drug-target interactions from heterogeneous biological data can advance our system-level search for drug molecules and therapeutic targets, which efforts have not yet reached full fruition. In this work, we report a systematic approach that efficiently integrates the chemical, genomic, and pharmacological information for drug targeting and discovery on a large scale, based on two powerful methods of Random Forest (RF) and Support Vector Machine (SVM). The performance of the derived models was evaluated and verified with internally five-fold cross-validation and four external independent validations. The optimal models show impressive performance of prediction for drug-target interactions, with a concordance of 82.83%, a sensitivity of 81.33%, and a specificity of 93.62%, respectively. The consistence of the performances of the RF and SVM models demonstrates the reliability and robustness of the obtained models. In addition, the validated models were employed to systematically predict known/unknown drugs and targets involving the enzymes, ion channels, GPCRs, and nuclear receptors, which can be further mapped to functional ontologies such as target-disease associations and target-target interaction networks. This approach is expected to help fill the existing gap between chemical genomics and network pharmacology and thus accelerate the drug discovery processes.     

Prediction of beta-turns with learning machines

The support vector machine approach was introduced to predict the beta-turns in proteins. The overall self-consistency rate by the re-substitution test for the training or learning dataset reached 100%. Both the training dataset and independent testing dataset were taken from Chou [J. Pept. Res. 49 (1997) 120]. The success prediction rates by the jackknife test for the beta-turn subset of 455 tetrapeptides and non-beta-turn subset of 3807 tetrapeptides in the training dataset were 58.1 and 98.4%, respectively. The success rates with the independent dataset test for the beta-turn subset of 110 tetrapeptides and non-beta-turn subset of 30,231 tetrapeptides were 69.1 and 97.3%, respectively. The results obtained from this study support the conclusion that the residue-coupled effect along a tetrapeptide is important for the formation of a beta-turn.     

Predicting novel substrates for enzymes with minimal experimental effort with active learning

Enzymatic substrate promiscuity is more ubiquitous than previously thought, with significant consequences for understanding metabolism and its application to biocatalysis. This realization has given rise to the need for efficient characterization of enzyme promiscuity. Enzyme promiscuity is currently characterized with a limited number of human-selected compounds that may not be representative of the enzyme's versatility. While testing large numbers of compounds may be impractical, computational approaches can exploit existing data to determine the most informative substrates to test next, thereby more thoroughly exploring an enzyme's versatility. To demonstrate this, we used existing studies and tested compounds for four different enzymes, developed support vector machine (SVM) models using these datasets, and selected additional compounds for experiments using an active learning approach. SVMs trained on a chemically diverse set of compounds were discovered to achieve maximum accuracies of ~80% using ~33% fewer compounds than datasets based on all compounds tested in existing studies. Active learning-selected compounds for testing resolved apparent conflicts in the existing training data, while adding diversity to the dataset. The application of these algorithms to wide arrays of metabolic enzymes would result in a library of SVMs that can predict high-probability promiscuous enzymatic reactions and could prove a valuable resource for the design of novel metabolic pathways.     

Strategies to search and design stabilizers of protein-protein interactions: a feasibility study

Since protein-protein interactions play a pivotal role in the communication on the molecular level in virtually every biological system and process, the search and design for modulators of such interactions is of utmost importance. In recent years many inhibitors for specific protein-protein interactions have been developed, however, in only a few cases, small and druglike molecules are able to interfere in the complex formation of proteins. On the other hand, there are several small molecules known to modulate protein-protein interactions by means of stabilizing an already assembled complex. To achieve this goal, a ligand is binding to a pocket, which is located rim-exposed at the interface of the interacting proteins, for example as the phytotoxin Fusicoccin, which stabilizes the interaction of plant H+-ATPase and 14-3-3 protein by nearly a factor of 100. To suggest alternative leads, we performed a virtual screening campaign to discover new molecules putatively stabilizing this complex. Furthermore, we screen a dataset of 198 transient recognition protein-protein complexes for cavities, which are located rim-exposed at their interfaces. We provide evidence for high similarity between such rim-exposed cavities and usual ligands accommodating active sites of enzymes. This analysis suggests that rim-exposed cavities at protein-protein interfaces are druggable binding sites. Therefore, the principle of stabilizing protein-protein interactions seems to be a promising alternative to the approach of the competitive inhibition of such interactions by small molecules.     

Predicting subcellular localization with AdaBoost Learner

Protein subcellular localization, which tells where a protein resides in a cell, is an important characteristic of a protein, and relates closely to the function of proteins. The prediction of their subcellular localization plays an important role in the prediction of protein function, genome annotation and drug design. Therefore, it is an important and challenging role to predict subcellular localization using bio-informatics approach. In this paper, a robust predictor, AdaBoost Learner is introduced to predict protein subcellular localization based on its amino acid composition. Jackknife cross-validation and independent dataset test were used to demonstrate that Adaboost is a robust and efficient model in predicting protein subcellular localization. As a result, the correct prediction rates were 74.98% and 80.12% for the Jackknife test and independent dataset test respectively, which are higher than using other existing predictors. An online server for predicting subcellular localization of proteins based on AdaBoost classifier was available on http://chemdata.shu. edu.cn/sl12.