Test Set R2 Research Articles

Investigation of inhibitory activity of monoamine oxidase A with 4D-QSAR using Fukui indices identifier

The inhibitory activity of monoamine oxidase A (MAO-A) was investigated as 4D- Quantitative Structure-Activity Relationship (4D-QSAR) using Molecular Conformer Electron Topological (MCET), a ligand-based method. The activity of MAO-A according to a 3D interaction was explained by a descriptor with the Fukui indices, which represents the hard-soft acid / base reactivity of atoms. The Fukui indices of atoms in clusters was used as the Local Reactive Descriptor (LRD) of the ligand side. Stochastic approaches were made for each sub-cluster proposed by the Genetic Algorithm (GA). The parameters at receptor side of pharmacophore (Pha) were calculated based on the Fukui indices used as descriptors on the ligand side. Pha was determined based on ligand of synthetic derivatives that are MAO-A inhibitors. Molecular conformers with the most suitable atoms with the template conformer can be selected as one of the most suitable spatial structures for interaction with the receptor. This serves to determine the LRD values ​​of the atoms in the conformer Pha. Depending on the values of the descriptors of the ligand at the points of interaction, it may be either an activity-increasing auxiliary groups (AG) or anti-pharmacophore shielding groups (APS). The model was developed with Leave One Out-Cross Validation (LOO-CV), for the 24 molecules in the training set, and was validated for 8 molecules in the test set. Inhibitory activities determined from the established model were compared with experimental results. For the training set and the external cross validation test set compounds, the Q2 (0.829) and R2 (0.818) of the statistical parameters were calculated, respectively.

Image-based QSAR Model for the Prediction of P-gp Inhibitory Activity of Epigallocatechin and Gallocatechin Derivatives.

Permeability glycoprotein (P-gp) is one of the cell membrane proteins that can push some drugs out of the cell causing drug tolerance and its inhibition can prevent drug resistance. In this study, we used image-based Quantitative Structure-Activity Relationship (QSAR) models to predict the P-gp inhibitory activity of epigallocatechin and gallocatechin derivatives. The 2D-chemical structures and their P-gp inhibitory activity were taken from literature. The pixels of images and their Principal Components (PCs) were calculated using MATLAB software. Principle Component Regression (PCR), Artificial Neural Network (ANN) and Support Vector Machine (SVM) approaches were used to develop QSAR models. Statistical parameters included the leave one out cross-validated correlation coefficient (q2) for internal validation of the models and R2 of test set, Root Mean Square Error (RMSE) and Concordance Correlation Coefficient (CCC) were applied for external validation. Six PCs from image analysis method were selected by stepwise regression for developing linear and non-linear models. Non-linear models i.e. ANN (with the R2 of 0.80 for test set) were chosen as the best for the established QSAR models. According to the result of the external validation, ANN model based on image analysis method can predict the P-gp inhibitory activity of epigallocatechin and gallocatechin derivatives better than the PCR and SVM models.

Using Machine Learning to Predict Enthalpy of Solvation

Enthalpy of solvation is an important thermodynamic property for studying molecular interactions. However, measuring enthalpies of solvation is non-trivial and time-consuming. Therefore, models that can predict enthalpy of solvation values are of significant worth to the general community. Here we present such models, based upon the Acree enthalpy of solvation open data dataset, which can be used to predict enthalpy of solvation values directly from structure. We created machine learning models for enthalpies of solvation in ethanol using open Chemistry Development Kit descriptors that have overall test-set R2 values of 0.89–0.90 and test-set root mean squared error values of 6.60–7.10 kJ·mol−1. The accuracy of the models was improved by limiting them to a single dominant cluster. Since our models were developed under Open Notebook Science conditions, they are fully reproducible and our techniques transparent and easily adaptable to other solvents.

Predictive QSAR modeling study on berberine derivatives with hypolipidemic activity.

Berberine (BBR), isolated from a Chinese herb, is identified as a new cholesterol-lowering small molecule, and hundreds of berberine derivatives have been obtained for optimization of their hypolipidemic activities in recent years. However, so far there is no available quantitative structure-activity relationship (QSAR) model used for the development of novel BBR analogues with hypolipidemic activities, mainly due to lack of lipid-lowering molecular mechanisms and target identification of BBR. In this paper, the tactics using ligand efficiency indice instead of pIC50 as the activity could be adopted for the development of BBR QSAR models. A series of 59 BBR derivatives with hypolipidemic activities have been studied and split randomly into three sets of training and test sets. Statistical quality of most building models shows obviously robust. Best calculated model that employs LLE indice as the activity (Model 6) has the following statistical parameters: for training set R2 =.984, Q2 =0.981, RMSE=0.1160, and for test set R2 =.989, RMSE=0.0067. This model would be used for the development of novel BBR analogues with lipid-lowering activities as a hit discovery tool.

Pharmacophore modelling, virtual screening and molecular docking studies on PLD1 inhibitors

Lipid metabolism plays a significant role in influenza virus replication and subsequent infection. The regulatory mechanism governing lipid metabolism and viral replication is not properly understood to date, but both Phospholipase D (PLD1 and PLD2) activities are stimulated in viral infection. In vitro studies indicate that chemical inhibition of PLD1 delays viral entry and reduction of viral loads. The current study reports a three-dimensional pharmacophore model based on 35 known PLD1 inhibitors. A sub-set of 25 compounds was selected as the training set and the remaining 10 compounds were kept in the test set. One hundred and twelve pharmacophore models were generated; a six-featured pharmacophore model (AADDHR.57) with survival score (2.69) produced a statistically significant three-dimensional quantitative structure–activity relationship model with r2 = 0.97 (internal training set), r2 = 0.71 (internal test set) and Q2 = 0.64. The predictive power of the pharmacophore model was validated with an external test set (r2 = 0.73) and a systematic virtual screening work-flow was employed showing an enrichment factor of 23.68 at the top 2% of the dataset (active and decoys). Finally, the model was used for screening of the filtered PubChem database to fetch molecules which can be proposed as potential PLD1 inhibitors for blocking influenza infection.

Consensus scoring model for the molecular docking study of mTOR kinase inhibitor

The discovery of mammalian target of rapamycin (mTOR) kinase inhibitors has always been a research hotspot of antitumor drugs. Consensus scoring used in the docking study of mTOR kinase inhibitors usually improves hit rate of virtual screening. Herein, we attempt to build a series of consensus scoring models based on a set of the common scoring functions. In this paper, twenty-five kinds of mTOR inhibitors (16 clinical candidate compounds and 9 promising preclinical compounds) are carefully collected, and selected for the molecular docking study used by the Glide docking programs within the standard precise (SP) mode. The predicted poses of these ligands are saved, and revaluated by twenty-six available scoring functions, respectively. Subsequently, consensus scoring models are trained based on the obtained rescoring results by the partial least squares (PLS) method, and validated by Leave-one-out (LOO) method. In addition, three kinds of ligand efficiency indices (BEI, SEI, and LLE) instead of pIC50 as the activity could greatly improve the statistical quality of build models. Two best calculated models 10 and 22 using the same BEI indice have following statistical parameters, respectively: for model 10, training set R2=0.767, Q2=0.647, RMSE=0.024, and for test set R2=0.932, RMSE=0.026; for model 22, raining set R2=0.790, Q2=0.627, RMSE=0.023, and for test set R2=0.955, RMSE=0.020. These two consensus scoring model would be used for the docking virtual screening of novel mTOR inhibitors.

Development of a Web-Enabled SVR-Based Machine Learning Platform and its Application on Modeling Transgene Expression Activity of Aminoglycoside-Derived Polycations

Support Vector Regression (SVR) has become increasingly popular in cheminformatics modeling. As a result, SVR-based machine learning algorithms, including Fuzzy-SVR and Least Square-SVR (LS-SVR) have been developed and applied in various research areas. However, at present, few downloadable packages or public-domain software are available for these algorithms. To address this need, we developed the Support vector regression-based Online Learning Equipment (SOLE) web tool (available at http://reccr.chem.rpi.edu/SOLE/index.html) as an online learning system to support predictive cheminformatics and materials informatics studies. In this work, we employed the SOLE system to model transgene expression efficacy of polymers obtained from aminoglycoside antibiotics, which allowed the results of several modeling approaches to be easily compared. All models had test set r2 of 0.96-0.98 and test set R2 of 0.79-0.84. Y-scrambling test showed the models were stable and not over-fitted. SOLE has a user-friendly interface and includes routine elements of performing QSAR/QSPR studies that can be applied in various research areas. It utilizes rational and sophisticated feature selection, model selection and model evaluation processes.

Ultrahigh resolution total organic carbon analysis using Fourier Transform Near Infrarred Reflectance Spectroscopy (FT‐NIRS)

Fourier transform near infrared reflectance spectroscopy (FT‐NIRS) is a cheap, rapid, and nondestructive method for analyzing organic sediment components. Here, we examine the robustness of a within lake FT‐NIRS calibration using a data set of almost 400 core samples from Lake Suigetsu, Japan, as a means to rapidly reconstruct % total organic carbon (TOC). We evaluate the best spectra pretreatment, examine different statistical approaches, and provide recommendations for the optimum number of calibration samples required for accurate predictions. Results show that the most robust method is based on first‐order derivatives of all spectra modeled with partial least squares regression. We construct a TOC model training set using 247 samples and a validation test set using 135 samples (for test set R2 = 0.951, RMSE = 0.280) to determine TOC and illustrate the use of the model in an ultrahigh resolution (e.g., 1 mm/annual) study of a long sediment core from a climatically sensitive archive.

Modeling of LIM-Kinase 2 Inhibitory Activity of Pyrrolopyrimidine Analogues: Useful in Treatment of Ocular Hypertension and Glaucoma

The LIM-Kinase 2 (LIMK2) inhibitory activity of a series of pyrrolopyrimidine analogs has been analyzed through combinatorial protocol in multiple linear regressions (CP-MLR) and partial least square (PLS) using different descriptors obtained from DRAGON software. The empirical, topological and charge descriptors have led to statistically significant QSAR models and showed good external predictivity as reflected in test set R2 values (0.782 to 0.888). The obtained structure-activity correlations underlined the significance of bulkiness and molecular polarizability in improving the activity. The topological descriptors suggested that open chain or branched substituents are favorable while cyclic /ring substituents are unfavorable for the activity. The descriptors identified in the study showed that pyrrolopyrimidine scaffold holds scope for modulating LIMK2 inhibitory activity. The study gives a direction for further exploration of chemical space of pyrrolopyrimidine analogs as LIMK2 inhibitors.

2-D QSAR model development for α-amino suberic acid derivatives as a novel anticancer agent

Structure activity relationship has been reported for 34 compounds derived from α-amino suberic acid. Compounds in this class typically inhibit human histone deacetylase by hyperacetylation of histone protein in both normal and cancer cells induce expression of p21 and differentiate surviving cancer cells to a non-proliferating phenotype. In the present study, quantitative structure activity relationship study was performed on a series of novel α-amino suberic acid derivatives as inhibitors of histone deacetylase. Multiple linear regression analysis was performed to derive quantitative structure activity relationship models which were further evaluated internally as well as externally for the prediction of activity. The best quantitative structure activity relationship model for multiple linear relationship analysis was selected with a correlation coefficient (r 2) of 0.855, cross validation r 2 (CV) 0.809 and the standard error 0.287. Further partial least square and neural network analysis were also performed for partial least square analysis correlation coefficient (r 2) 0.876, CV r 2 0.756, standard error 0.287 for neural networking analysis R 2 of training set 0.861, R 2 of test set 0.661. All of them generated comparable results which prove that the model formed is sound and has good predictability.

Consensus features of CP-MLR and GA in modeling HIV-1 RT inhibitory activity of 4-benzyl/benzoylpyridin-2-one analogues

The HIV-1 reverse transcriptase (RT) inhibitory activity of benzyl/benzoylpyridinones is modeled with molecular features identified in combinatorial protocol in multiple linear regression (CP-MLR) and genetic algorithm (GA). Among the features, nDB and LogP are found to be the most influential descriptors to modulate the activity. Although the coefficient of nDB suggested in favor of benzylpyridinones skeleton, the coefficient of LogP suggested the favorability of hydrophilic nature in compounds for better activity. The partial least squares analysis of the descriptors common to CP-MLR and GA has displayed their predictivity over the total descriptors identified in both the approaches. The back-propagation artificial neural networks model from the five most significant common descriptors (nDB, T(O..O), MATS8e, LogP, and BELp4) has explained 93.2% variance in the HIV-1 RT activity of the training set compounds and showed a test set r2 of 0.89. The results suggest that the descriptors have the ability to identify the patterns in the compounds to predict potential analogues.

Multidimensional Quantitative Structure-Activity Relationships of Diacylhydrazine Toxicity to Lepidopteran and Coleopteran Insect Pests

Insecticidal activity indices, i.e., median lethal dose causing mortality in 50% of insects treated (pLD50) values, of 77 Diacylhydrazines (DAHs) were determined against larvae of the Beet Armyworm (Spodoptera exigua), the rice stem borer (Chilo suppressalis), and the Colorado Potato Beetle (Leptinotarsa decemlineata). A series of multidimensional Comparative Molecular Field Analysis (CoMFA) Quantitative Structure–Activity Relationship (QSAR) models was constructed for each type of insect toxicity, and these models were evaluated by q2 of 67-member training sets and r2 of ten-member test sets. Favored models yielded r2 values in the range of 0.77–0.88, q2 values in the range of 0.54–0.62 after consideration of outliers, test set r2 values in the range of 0.54–0.72, and external q2 values in the range of 0.31–0.58. The structure–activity relationships for each type of insect toxicity can be partially, but not entirely, rationalized in view of Ecdysone Receptor (EcR) LBD sequence alignments and putative contact residues relative to the known crystallized Heliothis virescens EcR LBD/BYI06830 holoreceptor. Alternate ligand poses or LBD remodeling may pertain across members of the DAH chemotype.

Novel, Customizable Scoring Functions, Parameterized Using N-PLS, for Structure-Based Drug Discovery

The ability to accurately predict biological affinity on the basis of in silico docking to a protein target remains a challenging goal in the CADD arena. Typically, "standard" scoring functions have been employed that use the calculated docking result and a set of empirical parameters to calculate a predicted binding affinity. To improve on this, we are exploring novel strategies for rapidly developing and tuning "customized" scoring functions tailored to a specific need. In the present work, three such customized scoring functions were developed using a set of 129 high-resolution protein-ligand crystal structures with measured Ki values. The functions were parametrized using N-PLS (N-way partial least squares), a multivariate technique well-known in the 3D quantitative structure-activity relationship field. A modest correlation between observed and calculated pKi values using a standard scoring function (r2 = 0.5) could be improved to 0.8 when a customized scoring function was applied. To mimic a more realistic scenario, a second scoring function was developed, not based on crystal structures but exclusively on several binding poses generated with the Flo+ docking program. Finally, a validation study was conducted by generating a third scoring function with 99 randomly selected complexes from the 129 as a training set and predicting pKi values for a test set that comprised the remaining 30 complexes. Training and test set r2 values were 0.77 and 0.78, respectively. These results indicate that, even without direct structural information, predictive customized scoring functions can be developed using N-PLS, and this approach holds significant potential as a general procedure for predicting binding affinity on the basis of in silico docking.

Application of Validated QSAR Models of D1Dopaminergic Antagonists for Database Mining

Rigorously validated quantitative structure-activity relationship (QSAR) models have been developed for 48 antagonists of the dopamine D1 receptor and applied to mining chemical datasets to discover novel potential antagonists. Several QSAR methods have been employed, including comparative molecular field analysis (CoMFA), simulated annealing-partial least squares (SA-PLS), k-nearest neighbor (kNN), and support vector machines (SVM). With the exception of CoMFA, these approaches employed 2D topological descriptors generated with the MolConnZ software package (EduSoft, LLC. MolconnZ, version 4.05; http://www.eslc.vabiotech.com/ [4.05], 2003). The original dataset was split into training and test sets to allow for external validation of each training set model. The resulting models were characterized by cross-validated R2 (q2) for the training set and predictive R2 values for the test set of (q2/R2) 0.51/0.47 for CoMFA, 0.7/0.76 for kNN, R2 for the training and test sets of 0.74/0.71 for SVM, and training set fitness and test set R2 values of 0.68/0.63 for SA-PLS. Validated QSAR models with R2 > 0.7, (i.e., kNN and SVM) were used to mine three publicly available chemical databases: the National Cancer Institute (NCI) database of ca. 250,000 compounds, the Maybridge Database of ca. 56,000 compounds, and the ChemDiv Database of ca. 450,000 compounds. These searches resulted in only 54 consensus hits (i.e., predicted active by all models); five of them were previously characterized as dopamine D1 ligands, but were not present in the original dataset. A small fraction of the purported D1 ligands did not contain a catechol ring found in all known dopamine full agonist ligands, suggesting that they may be novel structural antagonist leads. This study illustrates that the combined application of predictive QSAR modeling and database mining may provide an important avenue for rational computer-aided drug discovery.