Quantitative structure-property relationships in pharmaceutical research - Part 2

Temperature-Dependent Density and Viscosity Prediction for Hydrocarbons: Machine Learning and Molecular Dynamics Simulations

Machine learning-based predictive models allow rapid and reliable prediction of material properties and facilitate innovative materials design. Base oils used in the formulation of lubricant products are complex hydrocarbons of varying sizes and structure. This study developed Gaussian process regression-based models to accurately predict the temperature-dependent density and dynamic viscosity of 305 complex hydrocarbons. In our approach, strongly correlated/collinear predictors were trimmed, important predictors were selected by least absolute shrinkage and selection operator (LASSO) regularization and prior domain knowledge, hyperparameters were systematically optimized by Bayesian optimization, and the models were interpreted. The approach provided versatile and quantitative structure-property relationship (QSPR) models with relatively simple predictors for determining the dynamic viscosity and density of complex hydrocarbons at any temperature. In addition, we developed molecular dynamics simulation-based descriptors and evaluated the feasibility and versatility of dynamic descriptors from simulations for predicting the material properties. It was found that the models developed using a comparably smaller pool of dynamic descriptors performed similarly in predicting density and viscosity to models based on many more static descriptors. The best models were shown to predict density and dynamic viscosity with coefficient of determination (R2) values of 99.6% and 97.7%, respectively, for all data sets, including a test data set of 45 molecules. Finally, partial dependency plots (PDPs), individual conditional expectation (ICE) plots, local interpretable model-agnostic explanation (LIME) values, and trimmed model R2 values were used to identify the most important static and dynamic predictors of the density and viscosity.

Brazilin from Caesalpinia sappan L. as a Proprotein Convertase Subtilisin/Kexin Type 9 (PCSK9) Inhibitor: Pharmacophore-Based Virtual Screening, In Silico Molecular Docking, and In Vitro Studies

Background

Proprotein convertase subtilisin/kexin type 9 (PCSK9) is a crucial regulator of low-density lipoprotein cholesterol (LDL-c) levels, as it binds to and degrades the LDL receptor (LDLR) in the lysosome of hepatocytes. Elevated levels of PCSK9 have been linked to an increased LDL-c plasma levels, thereby increasing the risk of cardiovascular disease (CVD), making it an attractive target for therapeutic interventions. As a way to inhibit PCSK9 action, we searched for naturally derived small molecules which can block the binding of PCSK9 to the LDLR.

Methods

In this study, we carried out in silico studies which consist of virtual screening using an optimized pharmacophore model and molecular docking studies using Pyrx 0.98. Effects of the candidate compounds were evaluated using in vitro PCSK9-LDLR binding assays kit.

Results

Eleven natural compounds that bind to PCSK9 were virtually screened form HerbalDB database, including brazilin. Next, molecular docking studies using Pyrx 0.98 showed that brazilin had the highest binding affinity with PCSK9 at -9.0 (Kcal/mol), which was higher than that of the other ten compounds. Subsequent in vitro PCSK9-LDLR binding assays established that brazilin decreased the binding of PCSK9 to the EGF-A fragment of the LDLR in a dose-dependent manner, with an IC50 value of 2.19 μM.

Conclusion

We have identified brazilin, which is derived from the Caesalpinia sappan herb, which can act as a small molecule inhibitor of PCSK9. Our findings suggest that screening for small molecules that can block the interaction between PCSK9 and the LDLR in silico and in vitro may be a promising approach for developing novel lipid-lowering therapy.

PyL3dMD: Python LAMMPS 3D molecular descriptors package

Molecular descriptors characterize the biological, physical, and chemical properties of molecules and have long been used for understanding molecular interactions and facilitating materials design. Some of the most robust descriptors are derived from geometrical representations of molecules, called 3-dimensional (3D) descriptors. When calculated from molecular dynamics (MD) simulation trajectories, 3D descriptors can also capture the effects of operating conditions such as temperature or pressure. However, extracting 3D descriptors from MD trajectories is non-trivial, which hinders their wide use by researchers developing advanced quantitative-structure-property-relationship models using machine learning. Here, we describe a suite of open-source Python-based post-processing routines, called PyL3dMD, for calculating 3D descriptors from MD simulations. PyL3dMD is compatible with the popular simulation package LAMMPS and enables users to compute more than 2000 3D molecular descriptors from atomic trajectories generated by MD simulations. PyL3dMD is freely available via GitHub and can be easily installed and used as a highly flexible Python package on all major platforms (Windows, Linux, and macOS). A performance benchmark study used descriptors calculated by PyL3dMD to develop a neural network and the results showed that PyL3dMD is fast and efficient in calculating descriptors for large and complex molecular systems with long simulation durations. PyL3dMD facilitates the calculation of 3D molecular descriptors using MD simulations, making it a valuable tool for cheminformatics studies.

A review on machine learning algorithms for the ionic liquid chemical space

There are thousands of papers published every year investigating the properties and possible applications of ionic liquids. Industrial use of these exceptional fluids requires adequate understanding of their physical properties, in order to create the ionic liquid that will optimally suit the application. Computational property prediction arose from the urgent need to minimise the time and cost that would be required to experimentally test different combinations of ions. This review discusses the use of machine learning algorithms as property prediction tools for ionic liquids (either as standalone methods or in conjunction with molecular dynamics simulations), presents common problems of training datasets and proposes ways that could lead to more accurate and efficient models.

Melting point prediction of organic molecules by deciphering the chemical structure into a natural language

Establishing quantitative structure-property relationships for the rational design of small molecule drugs at the early discovery stage is highly desirable. Using natural language processing (NLP), we proposed a machine learning model to process the line notation of small organic molecules, allowing the prediction of their melting points. The model prediction accuracy benefits from training upon different canonicalized SMILES forms of the same molecules and does not decrease with increasing size, complexity, and structural flexibility. When a combination of two different canonicalized SMILES forms is used to train the model, the prediction accuracy improves. Largely distinguished from the previous fragment-based or descriptor-based models, the prediction accuracy of this NLP-based model does not decrease with increasing size, complexity, and structural flexibility of molecules. By representing the chemical structure as a natural language, this NLP-based model offers a potential tool for quantitative structure-property prediction for drug discovery and development.

Machine learning guided prediction of liquid chromatography-mass spectrometry ionization efficiency for genotoxic impurities in pharmaceutical products

The limitation and control of genotoxic impurities (GTIs) has continued to receive attention from pharmaceutical companies and authorities for several decades. Because GTIs have the ability to damage deoxyribonucleic acid (DNA) and the potential to cause cancer, low-level quantitation is required to protect patients. A quick and easy method of determining the liquid chromatography-mass spectrometry (LC/MS) conditions for high-sensitivity analysis of GTIs may prospectively accelerate pharmaceutical development. In this study, a quantitative structure-property relationship (QSPR) model was developed for predicting the ionization efficiency of compounds using liquid-chromatography-mass spectrometry (LC/MS) parameters and molecular descriptors. Before implementing the QSPR prediction model, linear regression analysis was performed to model the relationship between the ionization efficiency and the LC/MS parameters for each compound. Comparison of the predicted peak areas with the experimentally observed peak areas showed good agreement based on the coefficient of determination (R² > 0.96). The machine learning-based QSPR approach begins with computation of the molecular descriptors expressing the physicochemical properties of a compound, followed by a genetic algorithm-based feature selection. Linear and nonlinear regression were performed, and support vector machine (SVM) was selected as the best machine learning algorithm for the prediction. The SVM algorithm was developed and optimized using 1031 experimental data points for nine compounds, including well-known GTIs. Validation of the model by comparison of the predicted and observed relative ionization efficiencies (RIE) showed a high coefficient of determination (R² = 0.96) and low root mean squared error value (RMSE = 0.118). Finally, this established prediction model was applied to hydrophilic interaction liquid chromatography coupled with MS for a new compound in new mobile phase compositions and new MS parameter settings. The RMSE of the predicted versus observed RIE was 0.203. This prediction accuracy was sufficient to determine the starting point of the LC/MS method development. The methodology demonstrated in this study can be used to determine the LC/MS conditions for high sensitivity analysis of GTIs.

The Current Landscape of Novel Formulations and the Role of Mathematical Modeling in Their Development

Drug delivery is an integral part of the drug development process, influencing safety and efficacy of active pharmaceutical ingredients. The application of nanotechnology has enabled the discovery of novel formulations for numerous therapeutic purposes across multiple disease areas. However, evaluation of novel formulations in clinical scenarios is slow and hampered due to various ethical and logistical barriers. Computational models have the ability to integrate existing domain knowledge and mathematical correlations, to rationalize the feasibility of using novel formulations for safely enhancing drug delivery, identifying suitable candidates, and reducing the burden on preclinical and clinical studies. In this review, types of novel formulations and their application through several routes of administration and the use of modeling approaches that can find application in different stages of the novel formulation development process are discussed.

Quantitative Structure-Property Relationship Approach in Formulation Development: an Overview

Chemoinformatics is emerging as a new trend to set drug discovery which correlates the relationship between structure and biological functions. The main aim of chemoinformatics refers to analyzing the similarity among molecules, searching the molecules in the structural database, finding potential drug molecule and their property. One of the key fields in chemoinformatics is quantitative structure-property relationship (QSPR), which is an alternative process to predict the various physicochemical and biopharmaceutical properties. This methodology expresses molecules via various numerical values or properties (descriptors), which encodes the structural characteristics of molecules and further used to calculate physicochemical properties of the molecule. The established QSPR model could be used to predict the properties of compounds that have been measured or even have been unknown, which ultimately accelerates the development process of a new molecule or the product. The formulation characteristics (drug release, transportability, bioavailability) can be predicted with the integration of QSPR approach. Therefore, QSPR modeling is an emerging trend to skip conventional drug as well as formulation development process. The current review highlights the overall process involved in the application of the QSPR approach in formulation development.

Spectrophores as one-dimensional descriptors calculated from three-dimensional atomic properties: applications ranging from scaffold hopping to multi-target virtual screening

Spectrophores are novel descriptors that are calculated from the three-dimensional atomic properties of molecules. In our current implementation, the atomic properties that were used to calculate spectrophores include atomic partial charges, atomic lipophilicity indices, atomic shape deviations and atomic softness properties. This approach can easily be widened to also include additional atomic properties. Our novel methodology finds its roots in the experimental affinity fingerprinting technology developed in the 1990’s by Terrapin Technologies. Here we have translated it into a purely virtual approach using artificial affinity cages and a simplified metric to calculate the interaction between these cages and the atomic properties. A typical spectrophore consists of a vector of 48 real numbers. This makes it highly suitable for the calculation of a wide range of similarity measures for use in virtual screening and for the investigation of quantitative structure–activity relationships in combination with advanced statistical approaches such as self-organizing maps, support vector machines and neural networks. In our present report we demonstrate the applicability of our novel methodology for scaffold hopping as well as virtual screening.

QSPR modeling of detonation parameters and sensitivity of some energetic materials: DFT vs. PM3 calculations

The quantitative structure-property relationship (QSPR) methodology was applied to describe and seek the relationship between the structures and energetic properties (and sensitivity) for some common energy compounds. An extended series of structural and energetic descriptors was obtained with density functional theory (DFT) B3LYP and semi-empirical PM3 approaches. Results indicate that QSPR model constructed using quantum descriptors can be applied to verify the confidence of calculation results compared with experimental data. It can be extended to predict the properties of similar compounds.

Quantitative Structure-Activity/Property/Toxicity Relationships through Conceptual Density Functional Theory-Based Reactivity Descriptors

Developing effective structure-activity/property/toxicity relationships (QSAR/QSPR/QSTR) is very helpful in predicting biological activity, property, and toxicity of a given set of molecules. Regular change in these properties with the structural alteration is the main reason to obtain QSAR/QSPR/QSTR models. The advancement in making different QSAR/QSPR/QSTR models to describe activity, property, and toxicity of various groups of molecules is reviewed in this chapter. The successful implementation of Conceptual Density Functional Theory (CDFT)-based global as well as local reactivity descriptors in modeling effective QSAR/QSPR/QSTR is highlighted.

Development and experimental test of support vector machines virtual screening method for searching Src inhibitors from large compound libraries

BACKGROUND

Src plays various roles in tumour progression, invasion, metastasis, angiogenesis and survival. It is one of the multiple targets of multi-target kinase inhibitors in clinical uses and trials for the treatment of leukemia and other cancers. These successes and appearances of drug resistance in some patients have raised significant interest and efforts in discovering new Src inhibitors. Various in-silico methods have been used in some of these efforts. It is desirable to explore additional in-silico methods, particularly those capable of searching large compound libraries at high yields and reduced false-hit rates.

RESULTS

We evaluated support vector machines (SVM) as virtual screening tools for searching Src inhibitors from large compound libraries. SVM trained and tested by 1,703 inhibitors and 63,318 putative non-inhibitors correctly identified 93.53%~ 95.01% inhibitors and 99.81%~ 99.90% non-inhibitors in 5-fold cross validation studies. SVM trained by 1,703 inhibitors reported before 2011 and 63,318 putative non-inhibitors correctly identified 70.45% of the 44 inhibitors reported since 2011, and predicted as inhibitors 44,843 (0.33%) of 13.56M PubChem, 1,496 (0.89%) of 168 K MDDR, and 719 (7.73%) of 9,305 MDDR compounds similar to the known inhibitors.

CONCLUSIONS

SVM showed comparable yield and reduced false hit rates in searching large compound libraries compared to the similarity-based and other machine-learning VS methods developed from the same set of training compounds and molecular descriptors. We tested three virtual hits of the same novel scaffold from in-house chemical libraries not reported as Src inhibitor, one of which showed moderate activity. SVM may be potentially explored for searching Src inhibitors from large compound libraries at low false-hit rates.

Revisiting the general solubility equation: in silico prediction of aqueous solubility incorporating the effect of topographical polar surface area

The General Solubility Equation (GSE) is a QSPR model based on the melting point and log P of a chemical substance. It is used to predict the aqueous solubility of nonionizable chemical compounds. However, its reliance on experimentally derived descriptors, particularly melting point, limits its applicability to virtual compounds. The studies presented show that the GSE is able to predict, to within 1 log unit, the experimental aqueous solubility (log S) for 81% of the compounds in a data set of 1265 diverse chemical structures (-8.48 < log S < 1.58). However, the predictive ability of the GSE is reduced to 75% when applied to a subset of the data (1160 compounds -6.00 < log S < 0.00), which discounts those compounds occupying the sparsely populated regions of data space. This highlights how sparsely populated extremities of data sets can significantly skew results for linear regression-based models. Replacing the melting point descriptor of the GSE with a descriptor which accounts for topographical polar surface area (TPSA) produces a model of comparable quality to the GSE (the solubility of 81% of compounds in the full data set predicted accurately). As such, we propose an alternative simple model for predicting aqueous solubility which replaces the melting point descriptor of the GSE with TPSA and hence can be applied to virtual compounds. In addition, incorporating TPSA into the GSE in addition to log P and melting point gives a three descriptor model that improves accurate prediction of aqueous solubility over the GSE by 5.1% for the full and 6.6% for the reduced data set, respectively.

Molecular topology as a novel approach for drug discovery

INTRODUCTION

Molecular topology (MT) has emerged in recent years as a powerful approach for the in silico generation of new drugs. One key part of MT is that, in the process of drug design/discovery, there is no need for an explicit knowledge of a drug's mechanism of action unlike other drug discovery methods.

AREAS COVERED

In this review, the authors introduce the topic by explaining briefly the most common methodology used today in drug design/discovery and address the most important concepts of MT and the methodology followed (QSAR equations, LDA, etc.). Furthermore, the significant results achieved, from this approach, are outlined and discussed.

EXPERT OPINION

The results outlined herein can be explained by considering that MT represents a new paradigm in the field of drug design. This means that it is not only an alternative method to the conventional methods, but it is also independent, that is, it represents a pathway to connect directly molecular structure with the experimental properties of the compounds (particularly drugs). Moreover, the process can be realized also in the reverse pathway, that is, designing new molecules from their topological pattern, what opens almost limitless expectations in new drugs development, given that the virtual universe of molecules is much greater than that of the existing ones.

Ligand Classifier of Adaptively Boosting Ensemble Decision Stumps (LiCABEDS) and its application on modeling ligand functionality for 5HT-subtype GPCR families

Advanced high-throughput screening (HTS) technologies generate great amounts of bioactivity data, and this data needs to be analyzed and interpreted with attention to understand how these small molecules affect biological systems. As such, there is an increasing demand to develop and adapt cheminformatics algorithms and tools in order to predict molecular and pharmacological properties on the basis of these large data sets. In this manuscript, we report a novel machine-learning-based ligand classification algorithm, named Ligand Classifier of Adaptively Boosting Ensemble Decision Stumps (LiCABEDS), for data-mining and modeling of large chemical data sets to predict pharmacological properties in an efficient and accurate manner. The performance of LiCABEDS was evaluated through predicting GPCR ligand functionality (agonist or antagonist) using four different molecular fingerprints, including Maccs, FP2, Unity, and Molprint 2D fingerprints. Our studies showed that LiCABEDS outperformed two other popular techniques, classification tree and Naive Bayes classifier, on all four types of molecular fingerprints. Parameters in LiCABEDS, including the number of boosting iterations, initialization condition, and a "reject option" boundary, were thoroughly explored and discussed to demonstrate the capability of handling imbalanced data sets, as well as its robustness and flexibility. In addition, the detailed mathematical concepts and theory are also given to address the principle behind statistical prediction models. The LiCABEDS algorithm has been implemented into a user-friendly software package that is accessible online at http://www.cbligand.org/LiCABEDS/ .

Identification of small molecule aggregators from large compound libraries by support vector machines

Small molecule aggregators non-specifically inhibit multiple unrelated proteins, rendering them therapeutically useless. They frequently appear as false hits and thus need to be eliminated in high-throughput screening campaigns. Computational methods have been explored for identifying aggregators, which have not been tested in screening large compound libraries. We used 1319 aggregators and 128,325 non-aggregators to develop a support vector machines (SVM) aggregator identification model, which was tested by four methods. The first is five fold cross-validation, which showed comparable aggregator and significantly improved non-aggregator identification rates against earlier studies. The second is the independent test of 17 aggregators discovered independently from the training aggregators, 71% of which were correctly identified. The third is retrospective screening of 13M PUBCHEM and 168K MDDR compounds, which predicted 97.9% and 98.7% of the PUBCHEM and MDDR compounds as non-aggregators. The fourth is retrospective screening of 5527 MDDR compounds similar to the known aggregators, 1.14% of which were predicted as aggregators. SVM showed slightly better overall performance against two other machine learning methods based on five fold cross-validation studies of the same settings. Molecular features of aggregation, extracted by a feature selection method, are consistent with published profiles. SVM showed substantial capability in identifying aggregators from large libraries at low false-hit rates.

QSPR modeling of thermal stability of nitroaromatic compounds: DFT vs. AM1 calculated descriptors

The quantitative structure-property relationship (QSPR) methodology was applied to predict the decomposition enthalpies of 22 nitroaromatic compounds, used as indicators of thermal stability. An extended series of descriptors (constitutional, topological, geometrical charge related and quantum chemical) was calculated at two different levels of theory: density functional theory (DFT) and semi-empirical AM1 approaches. Reliable models have been developed for each level, leading to similar correlations between calculated and experimental data (R(2) > 0.98). Hence, both of them can be employed as screening tools for the prediction of thermal stability of nitroaromatic compounds. If using the AM1 model presents the advantage to be less time consuming, DFT allows the calculation of more accurate molecular quantum properties, e.g., conceptual DFT descriptors. In this study, our best QSPR model is based on such descriptors, providing more chemical comprehensive relationships with decomposition reactivity, a particularly complex property for the specific class of nitroaromatic compounds.

Virtual screening of Abl inhibitors from large compound libraries by support vector machines

Abl promotes cancers by regulating cell morphogenesis, motility, growth, and survival. Successes of several marketed and clinical trial Abl inhibitors against leukemia and other cancers and appearances of reduced efficacies and drug resistances have led to significant interest in and efforts for developing new Abl inhibitors. In silico methods of pharmacophore, fragment, and molecular docking have been used in some of these efforts. It is desirable to explore other in silico methods capable of searching large compound libraries at high yields and reduced false-hit rates. We evaluated support vector machines (SVM) as a virtual screening tool for searching Abl inhibitors from large compound libraries. SVM trained and tested by 708 inhibitors and 65,494 putative noninhibitors correctly identified 84.4 to 92.3% inhibitors and 99.96 to 99.99% noninhibitors in 5-fold cross validation studies. SVM trained by 708 pre-2008 inhibitors and 65 494 putative noninhibitors correctly identified 50.5% of the 91 inhibitors reported since 2008 and predicted as inhibitors 29,072 (0.21%) of 13.56M PubChem, 659 (0.39%) of 168K MDDR, and 330 (5.0%) of 6638 MDDR compounds similar to the known inhibitors. SVM showed comparable yields and substantially reduced false-hit rates against two similarity based and another machine learning VS methods based on the same training and testing data sets and molecular descriptors. These suggest that SVM is capable of searching Abl inhibitors from large compound libraries at low false-hit rates.

On the prediction of thermal stability of nitroaromatic compounds using quantum chemical calculations

This work presents a new approach to predict thermal stability of nitroaromatic compounds based on quantum chemical calculations and on quantitative structure-property relationship (QSPR) methods. The data set consists of 22 nitroaromatic compounds of known decomposition enthalpy (taken as a macroscopic property related to explosibility) obtained from differential scanning calorimetry. Geometric, electronic and energetic descriptors have been selected and computed using density functional theory (DFT) calculation to describe the 22 molecules. First approach consisted in looking at their linear correlations with the experimental decomposition enthalpy. Molecular weight, electrophilicity index, electron affinity and oxygen balance appeared as the most correlated descriptors (respectively R(2)=0.76, 0.75, 0.71 and 0.64). Then multilinear regression was computed with these descriptors. The obtained model is a six-parameter equation containing descriptors all issued from quantum chemical calculations. The prediction is satisfactory with a correlation coefficient R(2) of 0.91 and a predictivity coefficient R(cv)(2) of 0.84 using a cross validation method.

Evaluation of virtual screening performance of support vector machines trained by sparsely distributed active compounds

Virtual screening performance of support vector machines (SVM) depends on the diversity of training active and inactive compounds. While diverse inactive compounds can be routinely generated, the number and diversity of known actives are typically low. We evaluated the performance of SVM trained by sparsely distributed actives in six MDDR biological target classes composed of a high number of known actives (983-1645) of high, intermediate, and low structural diversity (muscarinic M1 receptor agonists, NMDA receptor antagonists, thrombin inhibitors, HIV protease inhibitors, cephalosporins, and renin inhibitors). SVM trained by regularly sparse data sets of 100 actives show improved yields at substantially reduced false-hit rates compared to those of published studies and those of Tanimoto-based similarity searching method based on the same data sets and molecular descriptors. SVM trained by very sparse data sets of 40 actives (2.4%-4.1% of the known actives) predicted 17.5-39.5%, 23.0-48.1%, and 70.2-92.4% of the remaining 943-1605 actives in the high, intermediate, and low diversity classes, respectively, 13.8-68.7% of which are outside the training compound families. SVM predicted 99.97% and 97.1% of the 9.997 M PUBCHEM and 167K remaining MDDR compounds as inactive and 2.6%-8.3% of the 19,495-38,483 MDDR compounds similar to the known actives as active. These suggest that SVM has substantial capability in identifying novel active compounds from sparse active data sets at low false-hit rates.

Quantitative structure-property relationship study for estimation of quantitative calibration factors of some organic compounds in gas chromatography

Quantitative structure-property relationship (QSPR) models have been used to predict and explain gas chromatographic data of quantitative calibration factors (f(M)). This method allows for the prediction of quantitative calibration factors in a variety of organic compounds based on their structures alone. Stepwise multiple linear regression (MLR) and non-linear radial basis function neural network (RBFNN) were performed to build the models. The statistical characteristics provided by multiple linear model (R2=0.927, RMS=0.073; AARD=6.34% for test set) indicated satisfactory stability and predictive ability, while the predictive ability of RBFNN model is somewhat superior (R2=0.959; RMS=0.0648; AARD=4.85% for test set). This QSPR approach can contribute to a better understanding of structural factors of the compounds responsible for quantitative analysis by gas chromatography, and can be useful in predicting the quantitative calibration factors of other compounds.

A support vector machines approach for virtual screening of active compounds of single and multiple mechanisms from large libraries at an improved hit-rate and enrichment factor

Support vector machines (SVM) and other machine-learning (ML) methods have been explored as ligand-based virtual screening (VS) tools for facilitating lead discovery. While exhibiting good hit selection performance, in screening large compound libraries, these methods tend to produce lower hit-rate than those of the best performing VS tools, partly because their training-sets contain limited spectrum of inactive compounds. We tested whether the performance of SVM can be improved by using training-sets of diverse inactive compounds. In retrospective database screening of active compounds of single mechanism (HIV protease inhibitors, DHFR inhibitors, dopamine antagonists) and multiple mechanisms (CNS active agents) from large libraries of 2.986 million compounds, the yields, hit-rates, and enrichment factors of our SVM models are 52.4-78.0%, 4.7-73.8%, and 214-10,543, respectively, compared to those of 62-95%, 0.65-35%, and 20-1200 by structure-based VS and 55-81%, 0.2-0.7%, and 110-795 by other ligand-based VS tools in screening libraries of >or=1 million compounds. The hit-rates are comparable and the enrichment factors are substantially better than the best results of other VS tools. 24.3-87.6% of the predicted hits are outside the known hit families. SVM appears to be potentially useful for facilitating lead discovery in VS of large compound libraries.

Machine learning approaches for predicting compounds that interact with therapeutic and ADMET related proteins

Computational methods for predicting compounds of specific pharmacodynamic and ADMET (absorption, distribution, metabolism, excretion and toxicity) property are useful for facilitating drug discovery and evaluation. Recently, machine learning methods such as neural networks and support vector machines have been explored for predicting inhibitors, antagonists, blockers, agonists, activators and substrates of proteins related to specific therapeutic and ADMET property. These methods are particularly useful for compounds of diverse structures to complement QSAR methods, and for cases of unavailable receptor 3D structure to complement structure-based methods. A number of studies have demonstrated the potential of these methods for predicting such compounds as substrates of P-glycoprotein and cytochrome P450 CYP isoenzymes, inhibitors of protein kinases and CYP isoenzymes, and agonists of serotonin receptor and estrogen receptor. This article is intended to review the strategies, current progresses and underlying difficulties in using machine learning methods for predicting these protein binders and as potential virtual screening tools. Algorithms for proper representation of the structural and physicochemical properties of compounds are also evaluated.

Inhibitory mode of N-phenyl-4-pyrazolo[1,5-b] pyridazin-3-ylpyrimidin-2-amine series derivatives against GSK-3: molecular docking and 3D-QSAR analyses

Glycogen synthase kinase 3 (GSK-3) inhibition is an important research topic because of its wide range of associated health implications. The interaction mode of a series of N-phenyl-4-pyrazolo[1,5-b]pyridazin-3-ylpyrimidin-2-amine compounds with human GSK-3 has been studied using molecular docking and 3D-QSAR approaches. In the 3D-QSAR studies, the molecular alignment and conformation determination are so important that they affect the success of a model. Flexible docking (AutoDock3.0.5) was used for the determination of 'active' conformation and molecular alignment. Comparative molecular field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA) were used to develop 3D-QSAR models of 80 N-phenyl-4-pyrazolo[1,5-b]pyridazin-3-ylpyrimidin-2-amine compounds. The r(2) values were 0.870 and 0.861 for CoMFA and CoMSIA models, respectively. The predictive ability of these models was validated by 10 compounds of the test set. Mapping these models back to the topology of the active site of GSK-3 led to a better understanding of the vital N-phenyl-4-pyrazolo[1,5-b]pyridazin-3-ylpyrimidin-2-amines-GSK-3 interactions. The results demonstrate that combination of ligand-based and receptor-based modeling is a powerful approach to build 3D-QSAR models. The interaction mode from this study may be helpful for the design of a novel inhibitor and guide the selection of candidate sites for further experimental studies on site-directed mutagenesis.

Prediction of physicochemical properties based on neural network modelling

The literature describing neural network modelling to predict physicochemical properties of organic compounds from the molecular structure is reviewed from the perspective of pharmaceutical research. The standard three-layer, feed-forward neural network is the technique most frequently used, although the use of other techniques is increasing. Various approaches to describe the molecular structure have been successfully used, including molecular fragments, topological indices, and descriptors calculated by semi-empirical quantum chemical methods. Some physicochemical properties, such as octanol-water partition coefficient, water solubility, boiling point and vapour pressure, have been modelled by several research groups over the years using different approaches and structurally diverse large training sets. The prediction accuracy of most models seems to be rather close to the performance of the experimental measurements, when the accuracy is assessed with a test set from the working database. Results with independent test sets have been less satisfactory. Implications of this problem are discussed.

Discovery informatics: its evolving role in drug discovery

Drug discovery and development is a highly complex process requiring the generation of very large amounts of data and information. Currently this is a largely unmet informatics challenge. The current approaches to building information and knowledge from large amounts of data has been addressed in cases where the types of data are largely homogeneous or at the very least well-defined. However, we are on the verge of an exciting new era of drug discovery informatics in which methods and approaches dealing with creating knowledge from information and information from data are undergoing a paradigm shift. The needs of this industry are clear: Large amounts of data are generated using a variety of innovative technologies and the limiting step is accessing, searching and integrating this data. Moreover, the tendency is to move crucial development decisions earlier in the discovery process. It is crucial to address these issues with all of the data at hand, not only from current projects but also from previous attempts at drug development. What is the future of drug discovery informatics? Inevitably, the integration of heterogeneous, distributed data are required. Mining and integration of domain specific information such as chemical and genomic data will continue to develop. Management and searching of textual, graphical and undefined data that are currently difficult, will become an integral part of data searching and an essential component of building information- and knowledge-bases.

Novel local (fragment-based) topological molecular descriptors for QSpr/QSAR and molecular design

Novel molecular descriptors based on local spectral moments of the bond matrix are defined. Mathematical expressions relating bond moments to linear combinations of structural fragments are derived. The novel descriptors are used to describe boiling points of alcohols producing a good QSPR model accounting for more than 98% of variance. A quantitative structure-reactivity model is obtained to predict the specific rate constant (log k) of the nucleophilic addition of mercaptoacetic acid to 2-furylethylene derivatives. The model accounts for more than 96% of the variance in log k. Two other models were also obtained by using molecular connectivity indices and total spectral moments of the bond matrix that account for <84% of the variance in this reactivity index. A model based on quantum chemical descriptors accounts for the same variance than that obtained with bond moments. The model based on local moments permitted to compute the contribution of different structural fragments to the reactivity, and a good relationship (r = 0.98) was obtained with these group contributions with Hammett sigma(p) constants for 21 groups.

QSAR studies applied to the prediction of antigen-antibody interaction kinetics as measured by BIACORE

The objective of this work was to investigate the potential of the quantitative structure-activity relationships (QSAR) approach for predictive modulation of molecular interaction kinetics. A multivariate QSAR approach involving modifications in peptide sequence and buffer composition was recently used in an attempt to predict the kinetics of peptide-antibody interactions as measured by BIACORE. Quantitative buffer-kinetics relationships (QBKR) and quantitative sequence-kinetics relationships (QSKR) models were developed. Their predictive capacity was investigated in this study by comparing predicted and observed kinetic dissociation parameters (k(d)) for new antigenic peptides, or in new buffers. The range of experimentally measured k(d) variations was small (300-fold), limiting the practical value of the approach for this particular interaction. However, the models were validated from a statistical point of view. In QSKR, the leave-one-out cross validation gave Q(2) = 0.71 for 24 peptides (all but one outlier), compared to 0.81 for 17 training peptides. A more precise model (Q(2) = 0.92) could be developed when removing sets of peptides sharing distinctive structural features, suggesting that different peptides use slightly different binding modes. All models share the most important factor and are informative for structure-kinetics relationships. In QBKR, the measured effect on k(d) of individual additives in the buffers was consistent with the effect predicted from multivariate buffers. Our results open new perspectives for the predictive optimization of interaction kinetics, with important implications in pharmacology and biotechnology.

Computation of the physio-chemical properties and data mining of large molecular collections

Very large data sets of molecules screened against a broad range of targets have become available due to the advent of combinatorial chemistry. This information has led to the realization that ADME (absorption, distribution, metabolism, and excretion) and toxicity issues are important to consider prior to library synthesis. Furthermore, these large data sets provide a unique and important source of information regarding what types of molecular shapes may interact with specific receptor or target classes. Thus, the requirement for rapid and accurate data mining tools became paramount. To address these issues Pharmacopeia, Inc. formed a computational research group, The Center for Informatics and Drug Discovery (CIDD).* In this review we cover the work done by this group to address both in silico ADME modeling and data mining issues faced by Pharmacopeia because of the availability of a large and diverse collection (over 6 million discrete compounds) of drug-like molecules. In particular, in the data mining arena we discuss rapid docking tools and how we employ them, and we describe a novel data mining tool based on a ID representation of a molecule followed by a molecular sequence alignment step. For the ADME area we discuss the development and application of absorption, blood-brain barrier (BBB) and solubility models. Finally, we summarize the impact the tools and approaches might have on the drug discovery process.