Feature selection and transduction for prediction of molecular bioactivity for drug design

MOTIVATION: In drug discovery a key task is to identify characteristics that separate active (binding) compounds from inactive (non-binding) ones. An automated prediction system can help reduce resources necessary to carry out this task. RESULTS: Two methods for prediction of molecular bioactivity for drug design are introduced and shown to perform well in a data set previously studied as part of the KDD (Knowledge Discovery and Data Mining) Cup 2001. The data is characterized by very few positive examples, a very large number of features (describing three-dimensional properties of the molecules) and rather different distributions between training and test data. Two techniques are introduced specifically to tackle these problems: a feature selection method for unbalanced data and a classifier which adapts to the distribution of the the unlabeled test data (a so-called transductive method). We show both techniques improve identification performance and in conjunction provide an improvement over using only one of the techniques. Our results suggest the importance of taking into account the characteristics in this data which may also be relevant in other problems of a similar type.     

Prediction methods and databases within chemoinformatics: emphasis on drugs and drug candidates

MOTIVATION: To gather information about available databases and chemoinformatics methods for prediction of properties relevant to the drug discovery and optimization process. RESULTS: We present an overview of the most important databases with 2-dimensional and 3-dimensional structural information about drugs and drug candidates, and of databases with relevant properties. Access to experimental data and numerical methods for selecting and utilizing these data is crucial for developing accurate predictive in silico models. Many interesting predictive methods for classifying the suitability of chemical compounds as potential drugs, as well as for predicting their physico-chemical and ADMET properties have been proposed in recent years. These methods are discussed, and some possible future directions in this rapidly developing field are described.     

Prediction of chemical-protein interactions: multitarget-QSAR versus computational chemogenomic methods

Elucidation of chemical-protein interactions (CPI) is the basis of target identification and drug discovery. It is time-consuming and costly to determine CPI experimentally, and computational methods will facilitate the determination of CPI. In this study, two methods, multitarget quantitative structure-activity relationship (mt-QSAR) and computational chemogenomics, were developed for CPI prediction. Two comprehensive data sets were collected from the ChEMBL database for method assessment. One data set consisted of 81?689 CPI pairs among 50?924 compounds and 136 G-protein coupled receptors (GPCRs), while the other one contained 43?965 CPI pairs among 23?376 compounds and 176 kinases. The range of the area under the receiver operating characteristic curve (AUC) for the test sets was 0.95 to 1.0 and 0.82 to 1.0 for 100 GPCR mt-QSAR models and 100 kinase mt-QSAR models, respectively. The AUC of 5-fold cross validation were about 0.92 for both 176 kinases and 136 GPCRs using the chemogenomic method. However, the performance of the chemogenomic method was worse than that of mt-QSAR for the external validation set. Further analysis revealed that there was a high false positive rate for the external validation set when using the chemogenomic method. In addition, we developed a web server named CPI-Predictor, , which is available for free. The methods and tool have potential applications in network pharmacology and drug repositioning.     

CScore: a simple yet effective scoring function for protein-ligand binding affinity prediction using modified CMAC learning architecture

Protein-ligand docking is a computational method to identify the binding mode of a ligand and a target protein, and predict the corresponding binding affinity using a scoring function. This method has great value in drug design. After decades of development, scoring functions nowadays typically can identify the true binding mode, but the prediction of binding affinity still remains a major problem. Here we present CScore, a data-driven scoring function using a modified Cerebellar Model Articulation Controller (CMAC) learning architecture, for accurate binding affinity prediction. The performance of CScore in terms of correlation between predicted and experimental binding affinities is benchmarked under different validation approaches. CScore achieves a prediction with R = 0.7668 and RMSE = 1.4540 when tested on an independent dataset. To the best of our knowledge, this result outperforms other scoring functions tested on the same dataset. The performance of CScore varies on different clusters under the leave-cluster-out validation approach, but still achieves competitive result. Lastly, the target-specified CScore achieves an even better result with R = 0.8237 and RMSE = 1.0872, trained on a much smaller but more relevant dataset for each target. The large dataset of protein-ligand complexes structural information and advances of machine learning techniques enable the data-driven approach in binding affinity prediction. CScore is capable of accurate binding affinity prediction. It is also shown that CScore will perform better if sufficient and relevant data is presented. As there is growth of publicly available structural data, further improvement of this scoring scheme can be expected.     

Spectral clustering of protein sequences

An important problem in genomics is automatically clustering homologous proteins when only sequence information is available. Most methods for clustering proteins are local, and are based on simply thresholding a measure related to sequence distance. We first show how locality limits the performance of such methods by analysing the distribution of distances between protein sequences. We then present a global method based on spectral clustering and provide theoretical justification of why it will have a remarkable improvement over local methods. We extensively tested our method and compared its performance with other local methods on several subsets of the SCOP (Structural Classification of Proteins) database, a gold standard for protein structure classification. We consistently observed that, the number of clusters that we obtain for a given set of proteins is close to the number of superfamilies in that set; there are fewer singletons; and the method correctly groups most remote homologs. In our experiments, the quality of the clusters as quantified by a measure that combines sensitivity and specificity was consistently better [on average, improvements were 84% over hierarchical clustering, 34% over Connected Component Analysis (CCA) (similar to GeneRAGE) and 72% over another global method, TribeMCL].     

Towards the Improved Discovery and Design of Functional Peptides: Common Features of Diverse Classes Permit Generalized Prediction of Bioactivity

The conventional wisdom is that certain classes of bioactive peptides have specific structural features that endow their particular functions. Accordingly, predictions of bioactivity have focused on particular subgroups, such as antimicrobial peptides. We hypothesized that bioactive peptides may share more general features, and assessed this by contrasting the predictive power of existing antimicrobial predictors as well as a novel general predictor, PeptideRanker, across different classes of peptides.We observed that existing antimicrobial predictors had reasonable predictive power to identify peptides of certain other classes i.e. toxin and venom peptides. We trained two general predictors of peptide bioactivity, one focused on short peptides (4–20 amino acids) and one focused on long peptides ( amino acids). These general predictors had performance that was typically as good as, or better than, that of specific predictors. We noted some striking differences in the features of short peptide and long peptide predictions, in particular, high scoring short peptides favour phenylalanine. This is consistent with the hypothesis that short and long peptides have different functional constraints, perhaps reflecting the difficulty for typical short peptides in supporting independent tertiary structure.We conclude that there are general shared features of bioactive peptides across different functional classes, indicating that computational prediction may accelerate the discovery of novel bioactive peptides and aid in the improved design of existing peptides, across many functional classes. An implementation of the predictive method, PeptideRanker, may be used to identify among a set of peptides those that may be more likely to be bioactive.     

Comparisons of predictive values of binary medical diagnostic tests for paired designs

Positive and negative predictive values of a diagnostic test are key clinically relevant measures of test accuracy. Surprisingly, statistical methods for comparing tests with regard to these parameters have not been available for the most common study design in which each test is applied to each study individual. In this paper, we propose a statistic for comparing the predictive values of two diagnostic tests using this paired study design. The proposed statistic is a score statistic derived from a marginal regression model and bears some relation to McNemar's statistic. As McNemar's statistic can be used to compare sensitivities and specificities of diagnostic tests, parameters that condition on disease status, our statistic can be considered as an analog of McNemar's test for the problem of comparing predictive values, parameters that condition on test outcome. We report on the results of a simulation study designed to examine the properties of this test under a variety of conditions. The method is illustrated with data from a study of methods for diagnosis of coronary artery disease.     

Genetic algorithms applied to multi-class prediction for the analysis of gene expression data

MOTIVATION: An important challenge in the use of large-scale gene expression data for biological classification occurs when the expression dataset being analyzed involves multiple classes. Key issues that need to be addressed under such circumstances are the efficient selection of good predictive gene groups from datasets that are inherently 'noisy', and the development of new methodologies that can enhance the successful classification of these complex datasets. METHODS: We have applied genetic algorithms (GAs) to the problem of multi-class prediction. A GA-based gene selection scheme is described that automatically determines the members of a predictive gene group, as well as the optimal group size, that maximizes classification success using a maximum likelihood (MLHD) classification method. RESULTS: The GA/MLHD-based approach achieves higher classification accuracies than other published predictive methods on the same multi-class test dataset. It also permits substantial feature reduction in classifier genesets without compromising predictive accuracy. We propose that GA-based algorithms may represent a powerful new tool in the analysis and exploration of complex multi-class gene expression data. AVAILABILITY: Supplementary information, data sets and source codes are available at http://www.omniarray.com/bioinformatics/GA.     

Modeling Dynamic Systems with Efficient Ensembles of Process-Based Models

Ensembles are a well established machine learning paradigm, leading to accurate and robust models, predominantly applied to predictive modeling tasks. Ensemble models comprise a finite set of diverse predictive models whose combined output is expected to yield an improved predictive performance as compared to an individual model. In this paper, we propose a new method for learning ensembles of process-based models of dynamic systems. The process-based modeling paradigm employs domain-specific knowledge to automatically learn models of dynamic systems from time-series observational data. Previous work has shown that ensembles based on sampling observational data (i.e., bagging and boosting), significantly improve predictive performance of process-based models. However, this improvement comes at the cost of a substantial increase of the computational time needed for learning. To address this problem, the paper proposes a method that aims at efficiently learning ensembles of process-based models, while maintaining their accurate long-term predictive performance. This is achieved by constructing ensembles with sampling domain-specific knowledge instead of sampling data. We apply the proposed method to and evaluate its performance on a set of problems of automated predictive modeling in three lake ecosystems using a library of process-based knowledge for modeling population dynamics. The experimental results identify the optimal design decisions regarding the learning algorithm. The results also show that the proposed ensembles yield significantly more accurate predictions of population dynamics as compared to individual process-based models. Finally, while their predictive performance is comparable to the one of ensembles obtained with the state-of-the-art methods of bagging and boosting, they are substantially more efficient.     

Three-dimensional quantitative structure-activity relationship analyses of piperidine-based CCR5 receptor antagonists

The CCR5 chemokine receptor has recently been found to play a crucial role in the viral entry stage of HIV infection and has therefore become an attractive potential target for anti-HIV therapeutics. On the other hand, the lack of CCR5 crystal structure data has impeded the development of structure-based CCR5 antagonist design. In this paper, we compare two three-dimensional Quantitative Structure-Activity Relationship (3D-QSAR) methods: Comparative Molecular Field Analysis (CoMFA) and Comparative Molecular Similarity Indices Analysis (CoMSIA) on a series of piperidine-based CCR5 antagonists as an alternative approach to investigate the interaction between CCR5 antagonists and their receptor. Superimposition of antagonist structures was performed using two alignment rules: atomic/centroid rms fit and rigid body field fit techniques. The 3D QSAR models were derived from a training set of 72 compounds, and were found to have predictive capability for a set of 19 holdout test compounds. The resulting contour maps produced by the best CoMFA and CoMSIA models were used to identify the structural features relevant to biological activity in this series of compounds. Further analyses of these interaction-field contour maps also showed a high level of internal consistency.     

Predicting linear B-cell epitopes using string kernels

The identification and characterization of B-cell epitopes play an important role in vaccine design, immunodiagnostic tests, and antibody production. Therefore, computational tools for reliably predicting linear B-cell epitopes are highly desirable. We evaluated Support Vector Machine (SVM) classifiers trained utilizing five different kernel methods using fivefold cross-validation on a homology-reduced data set of 701 linear B-cell epitopes, extracted from Bcipep database, and 701 non-epitopes, randomly extracted from SwissProt sequences. Based on the results of our computational experiments, we propose BCPred, a novel method for predicting linear B-cell epitopes using the subsequence kernel. We show that the predictive performance of BCPred (AUC = 0.758) outperforms 11 SVM-based classifiers developed and evaluated in our experiments as well as our implementation of AAP (AUC = 0.7), a recently proposed method for predicting linear B-cell epitopes using amino acid pair antigenicity. Furthermore, we compared BCPred with AAP and ABCPred, a method that uses recurrent neural networks, using two data sets of unique B-cell epitopes that had been previously used to evaluate ABCPred. Analysis of the data sets used and the results of this comparison show that conclusions about the relative performance of different B-cell epitope prediction methods drawn on the basis of experiments using data sets of unique B-cell epitopes are likely to yield overly optimistic estimates of performance of evaluated methods. This argues for the use of carefully homology-reduced data sets in comparing B-cell epitope prediction methods to avoid misleading conclusions about how different methods compare to each other. Our homology-reduced data set and implementations of BCPred as well as the APP method are publicly available through our web-based server, BCPREDS, at: http://ailab.cs.iastate.edu/bcpreds/.     

A new strategy for improved secondary screening and lead optimization using high-resolution SPR characterization of compound-target interactions

Biophysical label-free assays such as those based on SPR are essential tools in generating high-quality data on affinity, kinetic, mechanistic and thermodynamic aspects of interactions between target proteins and potential drug candidates. Here we show examples of the integration of SPR with bioinformatic approaches and mutation studies in the early drug discovery process. We call this combination 'structure-based biophysical analysis'. Binding sites are identified on target proteins using information that is either extracted from three-dimensional structural analysis (X-ray crystallography or NMR), or derived from a pharmacore model based on known binders. The binding site information is used for in silico screening of a large substance library (e.g. available chemical directory), providing virtual hits. The three-dimensional structure is also used for the design of mutants where the binding site has been impaired. The wild-type target and the impaired mutant are then immobilized on different spots of the sensor chip and the interactions of compounds with the wild-type and mutant are compared in order to identify selective binders for the binding site of the target protein. This method can be used as a cost-effective alternative to high-throughput screening methods in cases when detailed binding site information is available. Here, we present three examples of how this technique can be applied to provide invaluable data during different phases of the drug discovery process.     

Prediction-based structured variable selection through the receiver operating characteristic curves

In many clinical settings, a commonly encountered problem is to assess accuracy of a screening test for early detection of a disease. In these applications, predictive performance of the test is of interest. Variable selection may be useful in designing a medical test. An example is a research study conducted to design a new screening test by selecting variables from an existing screener with a hierarchical structure among variables: there are several root questions followed by their stem questions. The stem questions will only be asked after a subject has answered the root question. It is therefore unreasonable to select a model that only contains stem variables but not its root variable. In this work, we propose methods to perform variable selection with structured variables when predictive accuracy of a diagnostic test is the main concern of the analysis. We take a linear combination of individual variables to form a combined test. We then maximize a direct summary measure of the predictive performance of the test, the area under a receiver operating characteristic curve (AUC of an ROC), subject to a penalty function to control for overfitting. Since maximizing empirical AUC of the ROC of a combined test is a complicated nonconvex problem (Pepe, Cai, and Longton, 2006, Biometrics62, 221-229), we explore the connection between the empirical AUC and a support vector machine (SVM). We cast the problem of maximizing predictive performance of a combined test as a penalized SVM problem and apply a reparametrization to impose the hierarchical structure among variables. We also describe a penalized logistic regression variable selection procedure for structured variables and compare it with the ROC-based approaches. We use simulation studies based on real data to examine performance of the proposed methods. Finally we apply developed methods to design a structured screener to be used in primary care clinics to refer potentially psychotic patients for further specialty diagnostics and treatment.     

Machine learning for target discovery in drug development

The discovery of macromolecular targets for bioactive agents is currently a bottleneck for the informed design of chemical probes and drug leads. Typically, activity profiling against genetically manipulated cell lines or chemical proteomics is pursued to shed light on their biology and deconvolute drug–target networks. By taking advantage of the ever-growing wealth of publicly available bioactivity data, learning algorithms now provide an attractive means to generate statistically motivated research hypotheses and thereby prioritize biochemical screens. Here, we highlight recent successes in machine intelligence for target identification and discuss challenges and opportunities for drug discovery.