Classification of kinase inhibitors using BCUT descriptors

Consensus holistic virtual screening for drug discovery: a novel machine learning model approach

In drug discovery, virtual screening is crucial for identifying potential hit compounds. This study aims to present a novel pipeline that employs machine learning models that amalgamates various conventional screening methods. A diverse array of protein targets was selected, and their corresponding datasets were subjected to active/decoy distribution analysis prior to scoring using four distinct methods: QSAR, Pharmacophore, docking, and 2D shape similarity, which were ultimately integrated into a single consensus score. The fine-tuned machine learning models were ranked using the novel formula "w_new", consensus scores were calculated, and an enrichment study was performed for each target. Distinctively, consensus scoring outperformed other methods in specific protein targets such as PPARG and DPP4, achieving AUC values of 0.90 and 0.84, respectively. Remarkably, this approach consistently prioritized compounds with higher experimental PIC50 values compared to all other screening methodologies. Moreover, the models demonstrated a range of moderate to high performance in terms of R2 values during external validation. In conclusion, this novel workflow consistently delivered superior results, emphasizing the significance of a holistic approach in drug discovery, where both quantitative metrics and active enrichment play pivotal roles in identifying the best virtual screening methodology.Scientific contributionWe presented a novel consensus scoring workflow in virtual screening, merging diverse methods for enhanced compound selection. We also introduced 'w_new', a groundbreaking metric that intricately refines machine learning model rankings by weighing various model-specific parameters, revolutionizing their efficacy in drug discovery in addition to other domains.

Binding Constants of Substituted Benzoic Acids with Bovine Serum Albumin

Experimental data on the affinity of various substances to albumin are essential for the development of empirical models to predict plasma binding of drug candidates. Binding of 24 substituted benzoic acid anions to bovine serum albumin was studied using spectrofluorimetric titration. The equilibrium constants of binding at 298 K were determined according to 1:1 complex formation model. The relationships between the ligand structure and albumin affinity are analyzed. The binding constant values for m- and p-monosubstituted acids show a good correlation with the Hammett constants of substituents. Two- and three-parameter quantitative structure–activity relationship (QSAR) models with theoretical molecular descriptors are able to satisfactorily describe the obtained values for the whole set of acids. It is shown that the electron-density distribution in the aromatic ring exerts crucial influence on the albumin affinity.

Development and Testing of Druglike Screening Libraries

Although significant advances in experimental high throughput screening (HTS) have been made for drug lead identification, in silico virtual screening (VS) is indispensable owing to its unique advantage over experimental HTS, target-focused, cheap, and efficient, albeit its disadvantage of producing false positive hits. For both experimental HTS and VS, the quality of screening libraries is crucial and determines the outcome of those studies. In this paper, we first reviewed the recent progress on screening library construction. We realized the urgent need for compiling high-quality screening libraries in drug discovery. Then we compiled a set of screening libraries from about 20 million druglike ZINC molecules by running fingerprint-based similarity searches against known drug molecules. Lastly, the screening libraries were objectively evaluated using 5847 external actives covering more than 2000 drug targets. The result of the assessment is very encouraging. For example, with the Tanimoto coefficient being set to 0.75, 36% of external actives were retrieved and the enrichment factor was 13. Additionally, drug target family specific screening libraries were also constructed and evaluated. The druglike screening libraries are available for download from https://mulan.pharmacy.pitt.edu .

Databases and QSAR for Cancer Research

In this review, we take a survey of bioinformatics databases and quantitative structure-activity relationship studies reported in published literature. Databases from the most general to special cancer-related ones have been included. Most commonly used methods of structure-based analysis of molecules have been reviewed, along with some case studies where they have been used in cancer research. This article is expected to be of use for general bioinformatics researchers interested in cancer and will also provide an update to those who have been actively pursuing this field of research.

Structure-based QSAR, molecule design and bioassays of protease-activated receptor 1 inhibitors

Quantitative structure-activity relationship (QSAR) studies were performed on a series of protease-activated receptor 1 (PAR1) inhibitors to identify the key structural features responsible for their biological activity. Induced-fit docking (IFD) was used to explore the active mechanisms of all PAR1 inhibitors at the active pocket of PAR1, and the best plausible conformation was determined by IFD for further QSAR studies. Based on the best plausible conformation, structure-based descriptors and ligand descriptors incorporating the ligand-receptor interaction were calculated. The random forest method was used to select important descriptors and build the 2D-QSAR model. The results of the 2D-QSAR model gave a squared correlation coefficient (R²) of 0.937, a prediction squared correlation coefficient (R²_pred) of 0.845 and a mean square error (MSE) of 0.056. Furthermore, a 3D-QSAR model was developed via topomer comparative molecular field analysis (Topomer CoMFA), resulting in an R² of 0.938, a cross-validated Q² of 0.503 and a R²_pred of 0.758. Based on the developed QSAR model, Topomer search was used for virtual screening of the R2 fragment in lead-like inhibitors from the National Cancer Institute (NCI) database, which contains 260,000 molecules. Eighty-two compounds were designed with different R2 fragments, and four of these compounds were selected for further biological testing. All four compounds showed inhibitory potency against PAR1.

SuperPred: update on drug classification and target prediction

The SuperPred web server connects chemical similarity of drug-like compounds with molecular targets and the therapeutic approach based on the similar property principle. Since the first release of this server, the number of known compound–target interactions has increased from 7000 to 665 000, which allows not only a better prediction quality but also the estimation of a confidence. Apart from the addition of quantitative binding data and the statistical consideration of the similarity distribution in all drug classes, new approaches were implemented to improve the target prediction. The 3D similarity as well as the occurrence of fragments and the concordance of physico-chemical properties is also taken into account. In addition, the effect of different fingerprints on the prediction was examined. The retrospective prediction of a drug class (ATC code of the WHO) allows the evaluation of methods and descriptors for a well-characterized set of approved drugs. The prediction is improved by 7.5% to a total accuracy of 75.1%. For query compounds with sufficient structural similarity, the web server allows prognoses about the medical indication area of novel compounds and to find new leads for known targets. SuperPred is publicly available without registration at: http://prediction.charite.de.

Compound acquisition and prioritization algorithm for constructing structurally diverse compound libraries

In the present study, we report a compound acquisition and prioritization algorithm established for rational chemical library purchasing or compound synthesis to increase the diversity of an existing compound collection. This method was established based on a chemistry-space calculation using BCUT (Burden CAS University of Texas) descriptors. To identify the acquisition of compounds from candidate collections into the existing collection, a derived distance-based selection rule was applied, and the results were well supported by pairwise similarity calculations and cell-partition statistics in chemistry space. The correlation between chemistry-space distance and Tanimoto similarity index was also studied to justify the compound acquisition strategy through weighted linear regression. As a rational approach for library design, the distance-based selection rule exhibits certain advantages in prioritizing compound selection to enhance the overall structural diversity of an existing in-house compound collection or virtual combinatorial library for in silico screening, diversity oriented synthesis, and high-throughput screening.

Methods for combinatorial and parallel library design

Diversity has historically played a critical role in design of combinatorial libraries, screening sets and corporate collections for lead discovery. Large library design dominated the field in the 1990s with methods ranging anywhere from purely arbitrary through property based reagent selection to product based approaches. In recent years, however, there has been a downward trend in library size. This was due to increased information about the desirable targets gleaned from the genomics revolution and to the ever growing availability of target protein structures from crystallography and homology modeling. Creation of libraries directed toward families of receptors such as GPCRs, kinases, nuclear hormone receptors, proteases, etc., replaced the generation of libraries based primarily on diversity while single target focused library design has remained an important objective. Concurrently, computing grids and cpu clusters have facilitated the development of structure based tools that screen hundreds of thousands of molecules. Smaller "smarter" combinatorial and focused parallel libraries replaced those early un-focused large libraries in the twenty-first century drug design paradigm. While diversity still plays a role in lead discovery, the focus of current library design methods has shifted to receptor based methods, scaffold hopping/bio-isostere searching, and a much needed emphasis on synthetic feasibility. Methods such as "privileged substructures based design" and pharmacophore based design still are important methods for parallel and small combinatorial library design. This chapter discusses some of the possible design methods and presents examples where they are available.

Lead Discovery Using Virtual Screening

The practice of virtual screening (VS) to identify chemical leads to known or novel targets is becoming a core function of the computational chemist within industry. By employing a range of techniques, when attempting to identify compounds with activity against a biological target, a small focused subset of a larger collection of compounds can be identified and tested, often with results much better than selecting a similar number of compounds at random. We will review the key methods available, their relative success, and provide practical insights into best practices and key gaps. We will also argue that the capability of VS methods has grown to a point where fuller integration with experimental methods, including HTS, could increase the effectiveness of both.

Developing and validating predictive decision tree models from mining chemical structural fingerprints and high-throughput screening data in PubChem

Background

Recent advances in high-throughput screening (HTS) techniques and readily available compound libraries generated using combinatorial chemistry or derived from natural products enable the testing of millions of compounds in a matter of days. Due to the amount of information produced by HTS assays, it is a very challenging task to mine the HTS data for potential interest in drug development research. Computational approaches for the analysis of HTS results face great challenges due to the large quantity of information and significant amounts of erroneous data produced.

Results

In this study, Decision Trees (DT) based models were developed to discriminate compound bioactivities by using their chemical structure fingerprints provided in the PubChem system . The DT models were examined for filtering biological activity data contained in four assays deposited in the PubChem Bioassay Database including assays tested for 5HT1a agonists, antagonists, and HIV-1 RT-RNase H inhibitors. The 10-fold Cross Validation (CV) sensitivity, specificity and Matthews Correlation Coefficient (MCC) for the models are 57.2~80.5%, 97.3~99.0%, 0.4~0.5 respectively. A further evaluation was also performed for DT models built for two independent bioassays, where inhibitors for the same HIV RNase target were screened using different compound libraries, this experiment yields enrichment factor of 4.4 and 9.7.

Conclusion

Our results suggest that the designed DT models can be used as a virtual screening technique as well as a complement to traditional approaches for hits selection.

Data mining a small molecule drug screening representative subset from NIH PubChem

PubChem is a scientific showcase of the NIH Roadmap Initiatives. It is a compound repository created to facilitate information exchange and data sharing among the NIH Roadmap-funded Molecular Library Screening Center Network (MLSCN) and the scientific community. However, PubChem has more than 10 million records of compound information. It will be challenging to conduct a drug screening of the whole database of millions of compounds. Thus, the purpose of the present study was to develop a data mining cheminformatics approach in order to construct a representative and structure-diverse sublibrary from the large PubChem database. In this study, a new chemical diverse representative subset, rePubChem, was selected by whole-molecule chemistry-space matrix calculation using the cell-based partition algorithm. The representative subset was generated and was then subjected to evaluations by compound property analyses based on 1D and 2D molecular descriptors. The new subset was also examined and assessed for self-similarity analysis based on 2D molecular fingerprints in comparing with the source compound library. The new subset has a much smaller library size (540K compounds) with minimum similarity and redundancy without loss of the structural diversity and basic molecular properties of its parent library (5.3 million compounds). The new representative subset library generated could be a valuable structure-diverse compound resource for in silico virtual screening and in vitro HTS drug screening. In addition, the established subset generation method of using the combined cell-based chemistry-space partition metrics with pairwised 2D fingerprint-based similarity search approaches will also be important to a broad scientific community interested in acquiring structurally diverse compounds for efficient drug screening, building representative virtual combinatorial chemistry libraries for syntheses, and data mining large compound databases like the PubChem library in general.

Assessing potency of c-Jun N-terminal kinase 3 (JNK3) inhibitors using 2D molecular descriptors and binary QSAR methodology

JNK3 signaling pathway is gaining interest due to its involvement in many neurological disorders. The purpose of this study was to explore for the first time the use of a large and diverse dataset in combination with binary QSAR methodology for predicting JNK3 activity class. Data were extracted from Aureus Pharma' AurSCOPE Kinase knowledge database and active or inactive classes were assigned to ligands based on IC50 biological activity. Two sets of 2D molecular descriptors (P_VSA and BCUT) were used to build models using different biological activity thresholds. The design of the models was preceded by the evaluation of the chemical space covered by the datasets and an assessment of its chemical diversity. The best model was found using a 100 nM IC50 threshold with surface-based P_VSA descriptors. This binary QSAR model reached an overall accuracy of 98% and a leave-one-out cross-validated accuracy of 94%. Most relevant descriptors were found to encode size and hydrophobic interactions. These derived models can be useful for screening chemical libraries in the search for new JNK3 inhibitors.

Introducing the consensus modeling concept in genetic algorithms: application to interpretable discriminant analysis

An evolutionary statistical learning method was applied to classify drugs according to their biological target and also to discriminate between a compilation of oral and nonoral drugs. The emphasis was placed not only on how well the models predict but also on their interpretability. In an enhancement to previous studies, the consistency of the model weights over several runs of the genetic algorithm was considered with the goal of producing comprehensible models. Via this approach, the descriptors and their ranges that contribute most to class discrimination were identified. Selecting a bin step size that enables the average descriptor properties of the class being trained to be captured improves the interpretability and discriminatory power of a model. The performance, consistency, and robustness of such models were further enhanced by using two novel approaches that reduce the variability between individual solutions: consensus and splice modeling. Finally, the ability of the genetic algorithm to discriminate between activity classes was compared with a similarity searching method, while naïve Bayes classifiers and support vector machines were applied in discriminating the oral and nonoral drugs.

Knowledge-based chemoinformatic approaches to drug discovery

The modern drug discovery process is steadily becoming more information driven. Structural, physicochemical and ADME-Tox property profiles of reference (successful) ligands, along with structural information of their target proteins, have been extremely useful for early-stage drug discovery. Recently, databases of known biologically active ligands (knowledge bases) have become more focused toward different protein-target classes. The number of new chemoinformatics tools used to analyze structures and properties of successful molecules has also increased enormously. Scientists in this area are exploring new physicochemical properties and appropriate drug sets to understand druglike properties. In this review, the various uses of the ligand knowledge bases in the drug discovery process have been critically reviewed.

Identifying Biologically Active Compound Classes Using Phenotypic Screening Data and Sampling Statistics

Scoring the activity of compounds in phenotypic high-throughput assays presents a unique challenge because of the limited resolution and inherent measurement error of these assays. Techniques that leverage the structural similarity of compounds within an assay can be used to improve the hit-recovery rate from screening data. A technique is presented that uses clustering and sampling statistics to predict likely compound activity by scoring entire structural classes. A set of phenotypic assays performed against a commercially available compound library was used as a test set. Using the class-scoring technique, the resultant activity prediction scores were more reproducible than individual assay measurements, and class scoring recovered known active compounds more efficiently than individual assay measurements because class scoring had fewer false positives. Known biologically active compounds were recovered 87% of the time using class scores, suggesting a low false-negative rate that compared well to individual assay measurements. In addition, many weak and potentially novel classes of active compounds, overlooked by individual assay measurements, were suggested.

Virtual screening of biogenic amine-binding G-protein coupled receptors: comparative evaluation of protein- and ligand-based virtual screening protocols

In this paper, we compare protein- and ligand-based virtual screening techniques for identifying the ligands of four biogenic amine-binding G-protein coupled receptors (GPCRs). For the screening of the virtual compound libraries, we used (1) molecular docking into GPCR homology models, (2) ligand-based pharmacophore and Feature Tree models, (3) three-dimensional (3D)-similarity searches, and (4) statistical methods [partial least squares (PLS) and partial least squares discriminant analysis (PLS-DA) models] based on two-dimensional (2D) molecular descriptors. The comparison of the different methods in retrieving known antagonists from virtual libraries shows that in our study the ligand-based pharmacophore-, Feature Tree-, and 2D quantitative structure-activity relationship (QSAR)-based screening techniques provide enrichment factors that are higher than those provided by molecular docking into the GPCR homology models. Nevertheless, the hit rates achieved when docking with GOLD and ranking the ligands with GoldScore (up to 60% among the top-ranked 1% of the screened databases) are still satisfying. These results suggest that docking into GPCR homology models can be a useful approach for lead finding by virtual screening when either little or no information about the active ligands is available.

The molecular basis for coxib inhibition of p38alpha MAP kinase

In this work, we present the results of two combined approaches, molecular docking and comparative molecular field analysis (CoMFA), to propose how the selective cyclooxygenase-2 inhibitor celecoxib could act as a p38 mitogen-activated protein (MAP) kinase inhibitor. The docking analysis revealed why celecoxib has a less favorable binding energy (DeltaG= -12.4kcal/mol) than the selective p38 MAP kinase (p38 MAPK) inhibitor, SB203580 (DeltaG= -22.2kcal/mol). The CoMFA results revealed unfavorable steric effects that can be related to the predicted lower p38 MAP kinase inhibitory activity of celecoxib. Additionally, FlexX and CoMFA results also suggested that etoricoxib, another selective COX-2 inhibitor, could inhibit p38 MAP kinase.

A 3D similarity method for scaffold hopping from known drugs or natural ligands to new chemotypes

A primary goal of 3D similarity searching is to find compounds with similar bioactivity to a reference ligand but with different chemotypes, i.e., "scaffold hopping". However, an adequate description of chemical structures in 3D conformational space is difficult due to the high-dimensionality of the problem. We present an automated method that simplifies flexible 3D chemical descriptions in which clustering techniques traditionally used in data mining are exploited to create "fuzzy" molecular representations called FEPOPS (feature point pharmacophores). The representations can be used for flexible 3D similarity searching given one or more active compounds without a priori knowledge of bioactive conformations or pharmacophores. We demonstrate that similarity searching with FEPOPS significantly enriches for actives taken from in-house high-throughput screening datasets and from MDDR activity classes COX-2, 5-HT3A, and HIV-RT, while also scaffold or ring-system hopping to new chemical frameworks. Further, inhibitors of target proteins (dopamine 2 and retinoic acid receptor) are recalled by FEPOPS by scaffold hopping from their associated endogenous ligands (dopamine and retinoic acid). Importantly, the method excels in comparison to commonly used 2D similarity methods (DAYLIGHT, MACCS, Pipeline Pilot fingerprints) and a commercial 3D method (Pharmacophore Distance Triplets) at finding novel scaffold classes given a single query molecule.

Hit-Directed Nearest-Neighbor Searching

This work describes a practical strategy used at Pharmacia for identifying compounds for follow-up screening following an initial HTS campaign against targets where no 3-D structural information is available and preliminary SAR models do not exist. The approach explicitly takes into account different representations of chemistry space and identifies compounds for follow-up screening that are likely to provide the best overall coverage of the chemistry spaces considered. Specifically, the work employs hit-directed nearest-neighbor (HDNN) searching of compound databases based upon a set of "probe compounds" obtained as hits in the preliminary high-throughput screens. Four different molecular representations that generate nearly unique chemistry spaces are used. The representations include 3-D, 2-D, 2-D topological BCUTs (2-DT) and molecular fingerprints derived from substructural fragments. In the case of the BCUT representations the NN searching is distance based, while in the case of molecular fingerprints a similarity-based measure is used. Generally, the results obtained differ significantly among all four methods, that is, the sets of NN compounds have surprisingly little overlap. Moreover, in all of the four chemistry space representations, a minimum of 3- to 4-fold enrichment in actives over random screening is observed even though the actives identified in each of the sets of NNs are in large measure unique. These results suggest that use of multiple searches based upon a variety of molecular representations provides an effective way of identifying more hits in HDNN searches of chemistry spaces than can be realized with single searches.

Classification of kinase inhibitors using a Bayesian model

The use of Bayesian statistics to model both general (multifamily) and specific (single-target) kinase inhibitors is investigated. The approach demonstrates an alternative to current computational methods applied to heterogeneous structure/activity data sets. This approach operates rapidly and is readily modifiable as required. A generalized model generated using inhibitor data from multiple kinase classes shows meaningful enrichment for several specific kinase targets. Such an approach can be used to prioritize compounds for screening or to optimally select compounds from third-party data collections. The observed benefit of the approach is finding compounds that are not structurally related to known actives, or novel targets for which there is not enough information to build a specific kinase model. The general kinase model described was built from a basis of mostly tyrosine kinase inhibitors, with some serine/threonine inhibitors; all the test cases used in prediction were also on tyrosine kinase targets. Confirming the applicability of this technique to other kinase families will be determined once those biological assays become available.

Selecting compounds for focused screening using linear discriminant analysis and artificial neural networks

Linear discriminant analysis and a committee of neural networks have been applied to recognise compounds that act at biological targets belonging to a specific gene family, protein kinases. The MDDR database was used to provide compounds targeted against this family and sets of randomly selected molecules. BCUT parameters were employed as input descriptors that encode structural properties and information relevant to ligand-receptor interactions. The technique was applied to purchasing compounds from external suppliers. These compounds achieved hit rates on a par with those achieved using known actives for related targets when tested for the ability to inhibit kinases at a single concentration. This approach is intended as one of a series of filters in the selection of screening candidates, compound purchases and the application of synthetic priorities to combinatorial libraries.

Further development of reduced graphs for identifying bioactive compounds

Reduced graphs provide summary representations of chemical structures. Here, a variety of different types of reduced graphs are compared in similarity searches. The reduced graphs are found to give comparable performance to Daylight fingerprints in terms of the number of active compounds retrieved. However, no one type of reduced graph is found to be consistently superior across a variety of different data sets. Consequently, a representative set of reduced graphs was chosen and used together with Daylight fingerprints in data fusion experiments. The results show improved performance in 10 out of 11 data sets compared to using Daylight fingerprints alone. Finally, the potential of using reduced graphs to build SAR models is demonstrated using recursive partitioning. An SAR model consistent with a published model is found following just two splits in the decision tree.

A consensus neural network-based technique for discriminating soluble and poorly soluble compounds

BCUT [Burden, CAS, and University of Texas] descriptors, defined as eigenvalues of modified connectivity matrices, have traditionally been applied to drug design tasks such as defining receptor relevant subspaces to assist in compound selections. In this paper we present studies of consensus neural networks trained on BCUTs to discriminate compounds with poor aqueous solubility from those with reasonable solubility. This level was set at 0.1 mg/mL on advice from drug formulation and drug discovery scientists. By applying strict criteria to the insolubility predictions, approximately 95% of compounds are classified correctly. For compounds whose predictions have a lower level of confidence, further parameters are examined in order to flag those considered to possess unsuitable biopharmaceutical and physicochemical properties. This approach is not designed to be applied in isolation but is intended to be used as a filter in the selection of screening candidates, compound purchases, and the application of synthetic priorities to combinatorial libraries.

Selecting screening candidates for kinase and G protein-coupled receptor targets using neural networks

A series of neural networks has been trained, using consensus methods, to recognize compounds that act at biological targets belonging to specific gene families. The MDDR database was used to provide compounds targeted against gene families and sets of randomly selected molecules. BCUT parameters were employed as input descriptors that encode structural properties and information relevant to ligand-receptor interactions. In each case, the networks identified over 80% of the compounds targeting a gene family. The technique was applied to purchasing compounds from external suppliers, and results from screening against one gene family demonstrated impressive abilities to predict the activity of the majority of known hit compounds.

The impact of informatics and computational chemistry on synthesis and screening

High-throughput synthesis and screening technologies have enhanced the impact of computational chemistry on the drug discovery process. From the design of targeted, drug-like libraries to 'virtual' optimization of potency, selectivity and ADME/Tox properties, computational chemists are able to efficiently manage costly resources and dramatically shorten drug discovery cycle times. This review will describe some of the successful strategies and applications of state-of-the-art algorithms to enhance drug discovery, as well as key points in the drug discovery process where computational methods can have, and have had, greatest impact.

The design of combinatorial libraries using properties and 3D pharmacophore fingerprints

Molecular diversity and similarity methods relevant to drug-receptor interactions are key for the design of combinatorial libraries for lead discovery and optimization. BCUT chemistry-space values for ligands have been used for many diversity-related applications and incorporate receptor-relevant properties such as hydrogen bonding and polarizability. Three-dimensional (3D) multiple-point pharmacophore descriptors (fingerprints) quantify diversity (or similarity) in terms of combinations of three or four functional groups associated with non-covalent ligand-receptor binding. BCUTs and pharmacophore fingerprints have been effectively utilized to design diversity libraries, and also show promise for focused library design.