Interaction model based on local protein substructures generalizes to the entire structural enzyme-ligand space

Sunsetting Binding MOAD with its last data update and the addition of 3D-ligand polypharmacology tools

Binding MOAD is a database of protein-ligand complexes and their affinities with many structured relationships across the dataset. The project has been in development for over 20 years, but now, the time has come to bring it to a close. Currently, the database contains 41,409 structures with affinity coverage for 15,223 (37%) complexes. The website BindingMOAD.org provides numerous tools for polypharmacology exploration. Current relationships include links for structures with sequence similarity, 2D ligand similarity, and binding-site similarity. In this last update, we have added 3D ligand similarity using ROCS to identify ligands which may not necessarily be similar in two dimensions but can occupy the same three-dimensional space. For the 20,387 different ligands present in the database, a total of 1,320,511 3D-shape matches between the ligands were added. Examples of the utility of 3D-shape matching in polypharmacology are presented. Finally, plans for future access to the project data are outlined.

DAMA: a method for computing multiple alignments of protein structures using local structure descriptors

MOTIVATION

The well-known fact that protein structures are more conserved than their sequences forms the basis of several areas of computational structural biology. Methods based on the structure analysis provide more complete information on residue conservation in evolutionary processes. This is crucial for the determination of evolutionary relationships between proteins and for the identification of recurrent structural patterns present in biomolecules involved in similar functions. However, algorithmic structural alignment is much more difficult than multiple sequence alignment. This study is devoted to the development and applications of DAMA-a novel effective environment capable to compute and analyze multiple structure alignments.

RESULTS

DAMA is based on local structural similarities, using local 3D structure descriptors and thus accounts for nearest-neighbor molecular environments of aligned residues. It is constrained neither by protein topology nor by its global structure. DAMA is an extension of our previous study (DEDAL) which demonstrated the applicability of local descriptors to pairwise alignment problems. Since the multiple alignment problem is NP-complete, an effective heuristic approach has been developed without imposing any artificial constraints. The alignment algorithm searches for the largest, consistent ensemble of similar descriptors. The new method is capable to capture most of the biologically significant similarities present in canonical test sets and is discriminatory enough to prevent the emergence of larger, but meaningless, solutions. Tests performed on the test sets, including protein kinases, demonstrate DAMA's capability of identifying equivalent residues, which should be very useful in discovering the biological nature of proteins similarity. Performance profiles show the advantage of DAMA over other methods, in particular when using a strict similarity measure QC, which is the ratio of correctly aligned columns, and when applying the methods to more difficult cases.

AVAILABILITY AND IMPLEMENTATION

DAMA is available online at http://dworkowa.imdik.pan.pl/EP/DAMA. Linux binaries of the software are available upon request.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Deciphering the synergistic network regulation of active components from SiNiSan against irritable bowel syndrome via a comprehensive strategy: Combined effects of synephrine, paeoniflorin and naringin

BACKGROUND

SiNiSan (SNS) is an ancient Chinese herbal prescription, and the current clinical treatment of irritable bowel syndrome (IBS) is effective. In the previous study of the research team, the multi-functional co-synergism of SNS against IBS was presented. Some potential drug targets and candidate ligands were predicted.

PURPOSE

This study attempts to explore the crucial ingredient combinations from SNS formula and reveal their synergistic mechanism for IBS therapy.

MATERIALS AND METHODS

In present study, a comprehensive strategy was performed to reveal IBS related pathways and biological modules, and explore synergistic effects of the ingredients, including ADME (absorption, distribution, metabolism, excretion) screening, Text mining, Venn analysis, Gene ontology (GO) analysis, Pathway cluster analysis, Molecular docking, Network construction and Experimental verification in visceral hypersensitivity (VHS) rats.

RESULTS

Three compressed IBS signal pathways were derived from ClueGO KEGG analysis of 63 IBS genes, including Neuroactive ligand-receptor interaction, Inflammatory mediator regulation of TRP (transient receptor potential) channels and Serotonergic synapse. A multi-module network, composed of four IBS therapeutic modules (psychological, inflammation, neuroendocrine and cross-talk modules), was revealed by Target-Pathway network. Nine kernel targets were considered closely associated with the IBS pathways, including ADRA2A, HTR2A, F2RL1, F2RL3, TRPV1, PKC, PKA, IL-1Β and NGF. In silico analysis revealed that three crucial ingredients (synephrine, paeoniflorin and naringin) were assumed to coordinate the network of those IBS therapeutic modules by acting on these kernel targets in the important pathways. In vivo experimental results showed that the crucial ingredient combinations synergistically affected the expressions of the kernel biological molecules, and improved the minimum capacity threshold of AWR in VHS rats.

CONCLUSION

The study proposes the important IBS associated pathways and the network regulation mechanisms of the crucial ingredients. It reveals the multi-target synergistic effect of the crucial ingredient combinations for the novel therapy on IBS.

The recent progress in proteochemometric modelling: focusing on target descriptors, cross-term descriptors and application scope

As an extension of the conventional quantitative structure activity relationship models, proteochemometric (PCM) modelling is a computational method that can predict the bioactivity relations between multiple ligands and multiple targets. Traditional PCM modelling includes three essential elements: descriptors (including target descriptors, ligand descriptors and cross-term descriptors), bioactivity data and appropriate learning functions that link the descriptors to the bioactivity data. Since its appearance, PCM modelling has developed rapidly over the past decade by taking advantage of the progress of different descriptors and machine learning techniques, along with the increasing amounts of available bioactivity data. Specifically, the new emerging target descriptors and cross-term descriptors not only significantly increased the performance of PCM modelling but also expanded its application scope from traditional protein-ligand interaction to more abundant interactions, including protein-peptide, protein-DNA and even protein-protein interactions. In this review, target descriptors and cross-term descriptors, as well as the corresponding application scope, are intensively summarized. Additionally, we look forward to seeing PCM modelling extend into new application scopes, such as Target-Catalyst-Ligand systems, with the further development of descriptors, machine learning techniques and increasing amounts of available bioactivity data.

Proteochemometric modeling of the antigen-antibody interaction: new fingerprints for antigen, antibody and epitope-paratope interaction

Despite the high specificity between antigen and antibody binding, similar epitopes can be recognized or cross-neutralized by paratopes of antibody with different binding affinities. How to accurately characterize this slight variation which may or may not change the antigen-antibody binding affinity is a key issue in this area. In this report, by combining cylinder model with shell structure model, a new fingerprint was introduced to describe both the structural and physical-chemical features of the antigen and antibody protein. Furthermore, beside the description of individual protein, the specific epitope-paratope interaction fingerprint (EPIF) was developed to reflect the bond and the environment of the antigen-antibody interface. Finally, Proteochemometric Modeling of the antigen-antibody interaction was established and evaluated on 429 antigen-antibody complexes. By using only protein descriptors, our model achieved the best performance (R2=0.91,Qtest2=0.68) among peers. Further, together with EPIF as a new cross-term, our model (R2=0.92,Qtest2=0.74) can significantly outperform peers with multiplication of ligand and protein descriptors as a cross-term (R2≤0.81,Qtest2≤0.44). Results illustrated that: 1) our newly designed protein fingerprints and EPIF can better describe the antigen-antibody interaction; 2) EPIF is a better and specific cross-term in Proteochemometric Modeling for antigen-antibody interaction. The fingerprints designed in this study will provide assistance to the description of antigen-antibody binding, and in future, it may be valuable help for the high-throughput antibody screening. The algorithm is freely available on request.

Synthetic biology for the directed evolution of protein biocatalysts: navigating sequence space intelligently

The amino acid sequence of a protein affects both its structure and its function. Thus, the ability to modify the sequence, and hence the structure and activity, of individual proteins in a systematic way, opens up many opportunities, both scientifically and (as we focus on here) for exploitation in biocatalysis. Modern methods of synthetic biology, whereby increasingly large sequences of DNA can be synthesised de novo, allow an unprecedented ability to engineer proteins with novel functions. However, the number of possible proteins is far too large to test individually, so we need means for navigating the 'search space' of possible protein sequences efficiently and reliably in order to find desirable activities and other properties. Enzymologists distinguish binding (Kd) and catalytic (kcat) steps. In a similar way, judicious strategies have blended design (for binding, specificity and active site modelling) with the more empirical methods of classical directed evolution (DE) for improving kcat (where natural evolution rarely seeks the highest values), especially with regard to residues distant from the active site and where the functional linkages underpinning enzyme dynamics are both unknown and hard to predict. Epistasis (where the 'best' amino acid at one site depends on that or those at others) is a notable feature of directed evolution. The aim of this review is to highlight some of the approaches that are being developed to allow us to use directed evolution to improve enzyme properties, often dramatically. We note that directed evolution differs in a number of ways from natural evolution, including in particular the available mechanisms and the likely selection pressures. Thus, we stress the opportunities afforded by techniques that enable one to map sequence to (structure and) activity in silico, as an effective means of modelling and exploring protein landscapes. Because known landscapes may be assessed and reasoned about as a whole, simultaneously, this offers opportunities for protein improvement not readily available to natural evolution on rapid timescales. Intelligent landscape navigation, informed by sequence-activity relationships and coupled to the emerging methods of synthetic biology, offers scope for the development of novel biocatalysts that are both highly active and robust.

Computational chemogenomics: is it more than inductive transfer?

High-throughput assays challenge us to extract knowledge from multi-ligand, multi-target activity data. In QSAR, weights are statically fitted to each ligand descriptor with respect to a single endpoint or target. However, computational chemogenomics (CG) has demonstrated benefits of learning from entire grids of data at once, rather than building target-specific QSARs. A possible reason for this is the emergence of inductive knowledge transfer (IT) between targets, providing statistical robustness to the model, with no assumption about the structure of the targets. Relevant protein descriptors in CG should allow one to learn how to dynamically adjust ligand attribute weights with respect to protein structure. Hence, models built through explicit learning (EL) by including protein information, while benefitting from IT enhancement, should provide additional predictive capability, notably for protein deorphanization. This interplay between IT and EL in CG modeling is not sufficiently studied. While IT is likely to occur irrespective of the injected target information, it is not clear whether and when boosting due to EL may occur. EL is only possible if protein description is appropriate to the target set under investigation. The key issue here is the search for evidence of genuine EL exceeding expectations based on pure IT. We explore the problem in the context of Support Vector Regression, using more than 9,400 pKi values of 31 GPCRs, where compound-protein interactions are represented by the concatenation of vectorial descriptions of compounds and proteins. This provides a unified framework to generate both IT-enhanced and potentially EL-enabled models, where the difference is toggled by supplied protein information. For EL-enabled models, protein information includes genuine protein descriptors such as typical sequence-based terms, but also the experimentally determined affinity cross-correlation fingerprints. These latter benchmark the expected behavior of a quasi-ideal descriptor capturing the actual functional protein-protein relatedness, and therefore thought to be the most likely to enable EL. EL- and IT-based methods were benchmarked alongside classical QSAR, with respect to cross-validation and deorphanization challenges. A rational method for projecting benchmarked methodologies into a strategy space is given, in the aims that the projection will provide directions for the types of molecule designs possible using a given methodology. While EL-enabled strategies outperform classical QSARs and favorably compare to similar published results, they are, in all respects evaluated herein, not strongly distinguished from IT-enhanced models. Moreover, EL-enabled strategies failed to prove superior in deorphanization challenges. Therefore, this paper raises caution that, contrary to common belief and intuitive expectation, the benefits of chemogenomics models over classical QSAR are quite possibly due less to the injection of protein-related information, and rather impacted more by the effect of inductive transfer, due to simultaneous learning from all of the modeled endpoints. These results show that the field of protein descriptor research needs further improvements to truly realize the expected benefit of EL.

Benchmarking of protein descriptor sets in proteochemometric modeling (part 2): modeling performance of 13 amino acid descriptor sets

Background

While a large body of work exists on comparing and benchmarking descriptors of molecular structures, a similar comparison of protein descriptor sets is lacking. Hence, in the current work a total of 13 amino acid descriptor sets have been benchmarked with respect to their ability of establishing bioactivity models. The descriptor sets included in the study are Z-scales (3 variants), VHSE, T-scales, ST-scales, MS-WHIM, FASGAI, BLOSUM, a novel protein descriptor set (termed ProtFP (4 variants)), and in addition we created and benchmarked three pairs of descriptor combinations. Prediction performance was evaluated in seven structure-activity benchmarks which comprise Angiotensin Converting Enzyme (ACE) dipeptidic inhibitor data, and three proteochemometric data sets, namely (1) GPCR ligands modeled against a GPCR panel, (2) enzyme inhibitors (NNRTIs) with associated bioactivities against a set of HIV enzyme mutants, and (3) enzyme inhibitors (PIs) with associated bioactivities on a large set of HIV enzyme mutants.

Results

The amino acid descriptor sets compared here show similar performance (<0.1 log units RMSE difference and <0.1 difference in MCC), while errors for individual proteins were in some cases found to be larger than those resulting from descriptor set differences ( > 0.3 log units RMSE difference and >0.7 difference in MCC). Combining different descriptor sets generally leads to better modeling performance than utilizing individual sets. The best performers were Z-scales (3) combined with ProtFP (Feature), or Z-Scales (3) combined with an average Z-Scale value for each target, while ProtFP (PCA8), ST-Scales, and ProtFP (Feature) rank last.

Conclusions

While amino acid descriptor sets capture different aspects of amino acids their ability to be used for bioactivity modeling is still – on average – surprisingly similar. Still, combining sets describing complementary information consistently leads to small but consistent improvement in modeling performance (average MCC 0.01 better, average RMSE 0.01 log units lower). Finally, performance differences exist between the targets compared thereby underlining that choosing an appropriate descriptor set is of fundamental for bioactivity modeling, both from the ligand- as well as the protein side.

Benchmarking of protein descriptor sets in proteochemometric modeling (part 1): comparative study of 13 amino acid descriptor sets

Background

While a large body of work exists on comparing and benchmarking of descriptors of molecular structures, a similar comparison of protein descriptor sets is lacking. Hence, in the current work a total of 13 different protein descriptor sets have been compared with respect to their behavior in perceiving similarities between amino acids. The descriptor sets included in the study are Z-scales (3 variants), VHSE, T-scales, ST-scales, MS-WHIM, FASGAI and BLOSUM, and a novel protein descriptor set termed ProtFP (4 variants). We investigate to which extent descriptor sets show collinear as well as orthogonal behavior via principal component analysis (PCA).

Results

In describing amino acid similarities, MSWHIM, T-scales and ST-scales show related behavior, as do the VHSE, FASGAI, and ProtFP (PCA3) descriptor sets. Conversely, the ProtFP (PCA5), ProtFP (PCA8), Z-Scales (Binned), and BLOSUM descriptor sets show behavior that is distinct from one another as well as both of the clusters above. Generally, the use of more principal components (>3 per amino acid, per descriptor) leads to a significant differences in the way amino acids are described, despite that the later principal components capture less variation per component of the original input data.

Conclusion

In this work a comparison is provided of how similar (and differently) currently available amino acids descriptor sets behave when converting structure to property space. The results obtained enable molecular modelers to select suitable amino acid descriptor sets for structure-activity analyses, e.g. those showing complementary behavior.

Screening of selective histone deacetylase inhibitors by proteochemometric modeling

BACKGROUND

Histone deacetylase (HDAC) is a novel target for the treatment of cancer and it can be classified into three classes, i.e., classes I, II, and IV. The inhibitors selectively targeting individual HDAC have been proved to be the better candidate antitumor drugs. To screen selective HDAC inhibitors, several proteochemometric (PCM) models based on different combinations of three kinds of protein descriptors, two kinds of ligand descriptors and multiplication cross-terms were constructed in our study.

RESULTS

The results show that structure similarity descriptors are better than sequence similarity descriptors and geometry descriptors in the leftacterization of HDACs. Furthermore, the predictive ability was not improved by introducing the cross-terms in our models. Finally, a best PCM model based on protein structure similarity descriptors and 32-dimensional general descriptors was derived (R2 = 0.9897, Qtest2 = 0.7542), which shows a powerful ability to screen selective HDAC inhibitors.

CONCLUSIONS

Our best model not only predict the activities of inhibitors for each HDAC isoform, but also screen and distinguish class-selective inhibitors and even more isoform-selective inhibitors, thus it provides a potential way to discover or design novel candidate antitumor drugs with reduced side effect.

Proteochemometric modeling of the bioactivity spectra of HIV-1 protease inhibitors by introducing protein-ligand interaction fingerprint

HIV-1 protease is one of the main therapeutic targets in HIV. However, a major problem in treatment of HIV is the rapid emergence of drug-resistant strains. It should be particularly helpful to clinical therapy of AIDS if one method can be used to predict antivirus capability of compounds for different variants. In our study, proteochemometric (PCM) models were created to study the bioactivity spectra of 92 chemical compounds with 47 unique HIV-1 protease variants. In contrast to other PCM models, which used Multiplication of Ligands and Proteins Descriptors (MLPD) as cross-term, one new cross-term, i.e. Protein-Ligand Interaction Fingerprint (PLIF) was introduced in our modeling. With different combinations of ligand descriptors, protein descriptors and cross-terms, nine PCM models were obtained, and six of them achieved good predictive abilities (Q(2)(test)>0.7). These results showed that the performance of PCM models could be improved when ligand and protein descriptors were complemented by the newly introduced cross-term PLIF. Compared with the conventional cross-term MLPD, the newly introduced PLIF had a better predictive ability. Furthermore, our best model (GD & P & PLIF: Q(2)(test) = 0.8271) could select out those inhibitors which have a broad antiviral activity. As a conclusion, our study indicates that proteochemometric modeling with PLIF as cross-term is a potential useful way to solve the HIV-1 drug-resistant problem.

A novel method to compare protein structures using local descriptors

Background

Protein structure comparison is one of the most widely performed tasks in bioinformatics. However, currently used methods have problems with the so-called "difficult similarities", including considerable shifts and distortions of structure, sequential swaps and circular permutations. There is a demand for efficient and automated systems capable of overcoming these difficulties, which may lead to the discovery of previously unknown structural relationships.

Results

We present a novel method for protein structure comparison based on the formalism of local descriptors of protein structure - DEscriptor Defined Alignment (DEDAL). Local similarities identified by pairs of similar descriptors are extended into global structural alignments. We demonstrate the method's capability by aligning structures in difficult benchmark sets: curated alignments in the SISYPHUS database, as well as SISY and RIPC sets, including non-sequential and non-rigid-body alignments. On the most difficult RIPC set of sequence alignment pairs the method achieves an accuracy of 77% (the second best method tested achieves 60% accuracy).

Conclusions

DEDAL is fast enough to be used in whole proteome applications, and by lowering the threshold of detectable structure similarity it may shed additional light on molecular evolution processes. It is well suited to improving automatic classification of structure domains, helping analyze protein fold space, or to improving protein classification schemes. DEDAL is available online at http://bioexploratorium.pl/EP/DEDAL.

From in silico target prediction to multi-target drug design: current databases, methods and applications

Given the tremendous growth of bioactivity databases, the use of computational tools to predict protein targets of small molecules has been gaining importance in recent years. Applications span a wide range, from the 'designed polypharmacology' of compounds to mode-of-action analysis. In this review, we firstly survey databases that can be used for ligand-based target prediction and which have grown tremendously in size in the past. We furthermore outline methods for target prediction that exist, both based on the knowledge of bioactivities from the ligand side and methods that can be applied in situations when a protein structure is known. Applications of successful in silico target identification attempts are discussed in detail, which were based partly or in whole on computational target predictions in the first instance. This includes the authors' own experience using target prediction tools, in this case considering phenotypic antibacterial screens and the analysis of high-throughput screening data. Finally, we will conclude with the prospective application of databases to not only predict, retrospectively, the protein targets of a small molecule, but also how to design ligands with desired polypharmacology in a prospective manner.

Proteochemometric modeling as a tool to design selective compounds and for extrapolating to novel targets

Proteochemometric modeling is founded on the principles of QSAR but is able to benefit from additional information in model training due to the inclusion of target information.

Towards Proteome-Wide Interaction Models Using the Proteochemometrics Approach

A proteochemometrics model was induced from all interaction data in the BindingDB database, comprizing in all 7078 protein-ligand complexes with representatives from all major drug target categories. Proteins were represented by alignment-independent sequence descriptors holding information on properties such as hydrophobicity, charge, and secondary structure. Ligands were represented by commonly used QSAR descriptors. The inhibition constant (pKi ) values of protein-ligand complexes were discretized into "high" and "low" interaction activity. Different machine-learning techniques were used to induce models relating protein and ligand properties to the interaction activity. The best was decision trees, which gave an accuracy of 80 % and an area under the ROC curve of 0.81. The tree pointed to the protein and ligand properties, which are relevant for the interaction. As the approach does neither require alignments nor knowledge of protein 3D structures virtually all available protein-ligand interaction data could be utilized, thus opening a way to completely general interaction models that may span entire proteomes.

Structure-Based Approaches to Target Fishing and Ligand Profiling

Chemogenomics is an emerging interdisciplinary field aiming at identifying all possible ligands of all possible targets. If one groups targets in columns and ligands in rows, chemogenomic approaches to drug discovery just fill the interaction matrix. Since experimental data do not suffice, several computational methods are currently actively developed to supplement time-consuming and costly experiments. They are either designed to fill rows and thus profile a ligand towards a heterogeneous set of targets (target profiling) or to fill columns and thus identify novel ligands for an existing target (standard virtual screening). At the interface of both strategies are now true chemogenomic computational methods filling well defined areas in the matrix. The present review will focus on (protein) structure-based approaches and illustrates major advances in this novel exciting field which is supposed to massively impact rational drug design in the next decade.

Binding affinity prediction with property-encoded shape distribution signatures

We report the use of the molecular signatures known as "property-encoded shape distributions" (PESD) together with standard support vector machine (SVM) techniques to produce validated models that can predict the binding affinity of a large number of protein ligand complexes. This "PESD-SVM" method uses PESD signatures that encode molecular shapes and property distributions on protein and ligand surfaces as features to build SVM models that require no subjective feature selection. A simple protocol was employed for tuning the SVM models during their development, and the results were compared to SFCscore, a regression-based method that was previously shown to perform better than 14 other scoring functions. Although the PESD-SVM method is based on only two surface property maps, the overall results were comparable. For most complexes with a dominant enthalpic contribution to binding (DeltaH/-TDeltaS > 3), a good correlation between true and predicted affinities was observed. Entropy and solvent were not considered in the present approach, and further improvement in accuracy would require accounting for these components rigorously.

A chemogenomics view on protein-ligand spaces

BACKGROUND

Chemogenomics is an emerging inter-disciplinary approach to drug discovery that combines traditional ligand-based approaches with biological information on drug targets and lies at the interface of chemistry, biology and informatics. The ultimate goal in chemogenomics is to understand molecular recognition between all possible ligands and all possible drug targets. Protein and ligand space have previously been studied as separate entities, but chemogenomics studies deal with large datasets that cover parts of the joint protein-ligand space. Since drug discovery has traditionally focused on ligand optimization, the chemical space has been studied extensively. The protein space has been studied to some extent, typically for the purpose of classification of proteins into functional and structural classes. Since chemogenomics deals not only with ligands but also with the macromolecules the ligands interact with, it is of interest to find means to explore, compare and visualize protein-ligand subspaces.

RESULTS

Two chemogenomics protein-ligand interaction datasets were prepared for this study. The first dataset covers the known structural protein-ligand space, and includes all non-redundant protein-ligand interactions found in the worldwide Protein Data Bank (PDB). The second dataset contains all approved drugs and drug targets stored in the DrugBank database, and represents the approved drug-drug target space. To capture biological and physicochemical features of the chemogenomics datasets, sequence-based descriptors were computed for the proteins, and 0, 1 and 2 dimensional descriptors for the ligands. Principal component analysis (PCA) was used to analyze the multidimensional data and to create global models of protein-ligand space. The nearest neighbour method, computed using the principal components, was used to obtain a measure of overlap between the datasets.

CONCLUSION

In this study, we present an approach to visualize protein-ligand spaces from a chemogenomics perspective, where both ligand and protein features are taken into account. The method can be applied to any protein-ligand interaction dataset. Here, the approach is applied to analyze the structural protein-ligand space and the protein-ligand space of all approved drugs and their targets. We show that this approach can be used to visualize and compare chemogenomics datasets, and possibly to identify cross-interaction complexes in protein-ligand space.

Integrating statistical predictions and experimental verifications for enhancing protein-chemical interaction predictions in virtual screening

Predictions of interactions between target proteins and potential leads are of great benefit in the drug discovery process. We present a comprehensively applicable statistical prediction method for interactions between any proteins and chemical compounds, which requires only protein sequence data and chemical structure data and utilizes the statistical learning method of support vector machines. In order to realize reasonable comprehensive predictions which can involve many false positives, we propose two approaches for reduction of false positives: (i) efficient use of multiple statistical prediction models in the framework of two-layer SVM and (ii) reasonable design of the negative data to construct statistical prediction models. In two-layer SVM, outputs produced by the first-layer SVM models, which are constructed with different negative samples and reflect different aspects of classifications, are utilized as inputs to the second-layer SVM. In order to design negative data which produce fewer false positive predictions, we iteratively construct SVM models or classification boundaries from positive and tentative negative samples and select additional negative sample candidates according to pre-determined rules. Moreover, in order to fully utilize the advantages of statistical learning methods, we propose a strategy to effectively feedback experimental results to computational predictions with consideration of biological effects of interest. We show the usefulness of our approach in predicting potential ligands binding to human androgen receptors from more than 19 million chemical compounds and verifying these predictions by in vitro binding. Moreover, we utilize this experimental validation as feedback to enhance subsequent computational predictions, and experimentally validate these predictions again. This efficient procedure of the iteration of the in silico prediction and in vitro or in vivo experimental verifications with the sufficient feedback enabled us to identify novel ligand candidates which were distant from known ligands in the chemical space.