Conformal prediction of HDAC inhibitors.

The growing interest in epigenetic probes and drug discovery, as revealed by several epigenetic drugs in clinical use or in the lineup of the drug development pipeline, is boosting the generation of screening data. In order to maximize the use of structure-activity relationships there is a clear need to develop robust and accurate models to understand the underlying structure-activity relationship. Similarly, accurate models should be able to guide the rational screening of compound libraries. Herein we introduce a novel approach for epigenetic quantitative structure-activity relationship (QSAR) modelling using conformal prediction. As a case study, we discuss the development of models for 11 sets of inhibitors of histone deacetylases (HDACs), which are one of the major epigenetic target families that have been screened. It was found that all derived models, for every HDAC endpoint and all three significance levels, are valid with respect to predictions for the external test sets as well as the internal validation of the corresponding training sets. Furthermore, the efficiencies for the predictions are above 80% for most data sets and above 90% for four data sets at different significant levels. The findings of this work encourage prospective applications of conformal prediction for other epigenetic target data sets.

A large comparison of integrated SAR/QSAR models of the Ames test for mutagenicity$.

Results from the Ames test are the first outcome considered to assess the possible mutagenicity of substances. Many QSAR models and structural alerts are available to predict this endpoint. From a regulatory point of view, the recommendation from international authorities is to consider the predictions of more than one model and to combine results in order to develop conclusions about the mutagenicity risk posed by chemicals. However, the results of those models are often conflicting, and the existing inconsistency in the predictions requires intelligent strategies to integrate them. In our study, we evaluated different strategies for combining results of models for Ames mutagenicity, starting from a set of 10 diverse individual models, each built on a dataset of around 6000 compounds. The novelty of our study is that we collected a much larger set of about 18,000 compounds and used the new data to build a family of integrated models. These integrations used probabilistic approaches, decision theory, machine learning, and voting strategies in the integration scheme. Results are discussed considering balanced or conservative perspectives, regarding the possible uses for different purposes, including screening of large collection of substances for prioritization.

Conformal prediction to define applicability domain - A case study on predicting ER and AR binding.

A fundamental element when deriving a robust and predictive in silico model is not only the statistical quality of the model in question but, equally important, the estimate of its predictive boundaries. This work presents a new method, conformal prediction, for applicability domain estimation in the field of endocrine disruptors. The method is applied to binders and non-binders related to the oestrogen and androgen receptors. Ensembles of decision trees are used as statistical method and three different sets (dragon, rdkit and signature fingerprints) are investigated as chemical descriptors. The conformal prediction method results in valid models where there is an excellent balance in quality between the internally validated training set and the corresponding external test set, both in terms of validity and with respect to sensitivity and specificity. With this method the level of confidence can be readily altered by the user and the consequences thereof immediately inspected. Furthermore, the predictive boundaries for the derived models are rigorously defined by using the conformal prediction framework, thus no ambiguity exists as to the level of similarity needed for new compounds to be in or out of the predictive boundaries of the derived models where reliable predictions can be expected.

Structure-interaction relationships between the bile acid GCA and pharmaceuticals using multivariate data analysis and capillary electrophoresis.

Capillary electrophoresis (CE) has been used in an interaction study of 66 pharmaceutical compounds with the bile acid glycocholate (GCA). The developed method proved to have a high precision in its ability to determine the mobility of drugs in buffer and buffer bile acids solutions. The relationship between solute structure and interaction with GCA was studied using two-dimensional descriptors with the in-house software SELMA and a three-dimensional model (quantum mechanical descriptors) in combination with the experimental CE-interaction data. The multivariate analysis method used was projection to latent structures by means of partial least squares (PLS). Two selections of training and test set were used for evaluation of a two-class model on interaction data. In the first selection all observations were used for training set, for example, creating a model, and re-predicting the observations on the model. A successful prediction on 85% of the drugs was observed using this model. The second selection used the 21 first tested compounds in the training set, where 78% of the compounds were correctly predicted using the two-dimensional model (SELMA) on the remaining 45 compounds and, respectively, 82% using the three-dimensional (quantum mechanical) model. Analysis of the impact of the descriptors showed that descriptors relating to hydrophobicity have a large positive effect on the interaction. Descriptors relating to polar properties have a pronounced negative effect on the interaction of drugs with bile acids.

In silico modelling of ADMET-a minireview of work from 2000 to 2004.

This article represents a minireview of work published so far in the 21st century in the in silico ADMET field of research related to investigations in the areas of solubility, hERG and cytochrome P450 3A4. Various approaches including 2D- and 3D-QSARs and pharmacophore modelling are discussed. The pros and cons of the methods used and models derived are examined. More general remarks on model development and validation are also reported.

Theoretical calculation and prediction of drug transport processes using simple parameters and partial least squares projections to latent structures (PLS) statistics. The use of electrotopological state indices.

A method of modeling and predicting drug transport processes using simple, theoretically computed molecular descriptors and multivariate statistics has been investigated in four data sets related to Caco-2 cell permeability, human intestinal absorption, brain-blood partitioning, and immobilized artificial membrane (IAM) chromatography. The program Molconn-Z was used to compute theoretical molecular descriptors related to electrotopological state indices. Additional parameters related to size and lipophilicity [i.e., calculated molar refraction (CMR) and octanol-water partition coefficient (CLOGP)] were also used in the statistical modeling. Good statistical models were derived (r(2) and Q(2) values ranged from 0.75 to 0.93 and 0.70 to 0.89, respectively) that permit fast computational (electronic) screening and prioritization of virtual libraries.

Experimental and computational screening models for the prediction of intestinal drug absorption.

The aim of this study was to devise experimental protocols and computational models for the prediction of intestinal drug permeability. Both the required experimental and computational effort and the accuracy and quality of the resulting predictions were considered. In vitro intestinal Caco-2 cell monolayer permeabilities were determined both in a highly accurate experimental setting (Pc) and in a faster, but less accurate, mode (Papp). Computational models were built using four different principles for generation of molecular descriptors (atom counts, molecular mechanics calculations, fragmental, and quantum mechanics approaches) and were evaluated for their ability to predict intestinal membrane permeability. A theoretical deconvolution of the polar molecular surface area (PSA) was also performed to facilitate the interpretation of this composite descriptor and allow the calculation of PSA in a simplified and fast mode. The results indicate that it is possible to predict intestinal drug permeability from rather simple models with little or no loss of accuracy. A new, fast computational model, based on partitioned molecular surface areas, that predicts intestinal drug permeability with an accuracy comparable to that of time-consuming quantum mechanics calculations is presented.

Prediction of drug transport processes using simple parameters and PLS statistics. The use of ACD/logP and ACD/ChemSketch descriptors.

A method of modelling and predicting biopharmaceutical properties using simple theoretically computed molecular descriptors and multivariate statistics has been investigated for several data sets related to solubility, IAM chromatography, permeability across Caco-2 cell monolayers, human intestinal perfusion, brain-blood partitioning, and P-glycoprotein ATPase activity. The molecular descriptors (e.g. molar refractivity, molar volume, index of refraction, surface tension and density) and logP were computed with ACD/ChemSketch and ACD/logP, respectively. Good statistical models were derived that permit simple computational prediction of biopharmaceutical properties. All final models derived had R(2) values ranging from 0.73 to 0.95 and Q(2) values ranging from 0.69 to 0.86. The RMSEP values for the external test sets ranged from 0.24 to 0.85 (log scale).

Prediction of polar surface area and drug transport processes using simple parameters and PLS statistics.

Modeling of the calculated polar surface area of drugs with rapidly derived descriptors (i.e., the number of hydrogen bonds accepting oxygen and nitrogen atoms and the number of hydrogen atoms bonded to these) using partial least squares projection to latent structures (PLS) analysis is described. The statistical analysis showed strong relationships between the hydrogen-bonding descriptors and the calculated polar surface area of five chemically diverse sets of drugs (R2>0.93 and Q2>0.69, n = 11, 20, 45, 70, and 74, respectively). The statistical models (using H-bonding descriptors and log P) of transport across Caco-2 cells (n = 11), brain-blood partitioning (two data sets, n = 45 and 70) and percent intestinal absorption (n = 20) showed R2 = 0.92, 0.72, 0.76, and 0.81 and Q2 = 0.74, 0.75, 0.71, and 0.73, respectively. The inclusion of log P improved two models, had no effect on one model, and had a slightly negative impact on one model. The combination of H-bonding descriptors with log P is similar to the Lipinski "rule-of-five" mnemonic. However, by using a multivariate statistical method (e.g., PLS), the prediction becomes quantitative instead of qualitative. Good statistical models were derived which permit fast computational screening and prioritization of virtual compound libraries.

Refinement of Catalyst hypotheses using simplex optimisation.

The program HypoOpt in combination with the MSI program citest has been used to optimise and expand 3D QSAR Catalyst hypotheses using simplex optimisation coupled with cross-validation. Three data sets related to angiotensin converting enzyme inhibition, squalene epoxidase inhibition and HIV protease inhibition were used to investigate the outcome of hypothesis optimisation. Simplex optimisation using leave-one-out cross-validation during the hypothesis refinement resulted in improved models with respect to predictivity of an external test set. Furthermore, the utilisation of the geometry of the active site for the HIV protease inhibitors, represented by Catalyst 'excluded volume' features, resulted in an optimised hypothesis with improved predictivity compared with the corresponding hypothesis derived without receptor information.

Theoretical calculation and prediction of P-glycoprotein-interacting drugs using MolSurf parametrization and PLS statistics.

A method for the modelling and prediction of P-glycoprotein-associated ATPase activity using theoretically computed molecular descriptors and multivariate statistics has been investigated using 22 diverse drug-like compounds. The program MolSurf was used to compute theoretical molecular descriptors related to physicochemical properties such as lipophilicity, polarity, polarizability and hydrogen bonding. The multivariate partial least squares projections to latent structures (PLS) method was used to delineate the relationship between the P-glycoprotein-associated ATPase activity and the theoretically computed molecular descriptors. The PLS analysis of the entire data set, with the exclusion of tamoxifen, resulted in one significant PLS component according to cross-validation with R(2)=0.718, Q(2)=0. 695, S.D.=0.475, F=48.37, RMSE(tr)=0.452, p<0.001. Properties associated with the size of the molecular surface, polarizability and hydrogen bonding had the largest impact on the P-glycoprotein-associated ATPase activity. All these properties should be high to promote high ATPase activity.

Beat-to-beat QRS amplitude variability in elite endurance athletes.

Results from variance electrocardiography, displaying the wide-band, phase-locked electrical micro-variability during the depolarization phase, was analysed versus clinical data, echocardiographic structural and functional variables and myocardial scintigraphic findings in 174 elite orienteers compared with 37 age-matched elite endurance athletes and 50 age-matched, healthy medical students. PCA analysis identified a subgroup of five orienteers deviating from the rest of the study group and both control groups with regard to their QRS amplitude variability. No correlations were found between pathology by medical history or any of the echocardiographic and scintigraphic variables or by the variance electrocardiographic aberrations in any of the groups studied.

Theoretical calculation and prediction of intestinal absorption of drugs in humans using MolSurf parametrization and PLS statistics.

A method for modeling and prediction of the intestinal absorption of drugs in humans using theoretically computed molecular descriptors and multivariate statistics has been investigated using 20 diverse drug-like compounds. The program MolSurf was used to compute theoretical molecular descriptors related to physicochemical properties such as lipophilicity, polarity, polarizability and hydrogen bonding. The multivariate Partial Least Squares Projections to Latent Structures (PLS) method was used to delineate the relationship between the intestinal absorption of drugs in humans and the theoretically computed molecular descriptors.Good statistical models were derived. Properties associated with hydrogen bonding had the largest impact on absorption and should be kept to a minimum to promote high absorption. High charge-transfer properties and the presence of surface electrons, i.e. valence electrons, which are not tightly bonded to the molecule, were also found to promote high absorption.

Theoretical calculation and prediction of brain-blood partitioning of organic solutes using MolSurf parametrization and PLS statistics.

Sixty-three structurally diverse compounds were investigated to statistically model the brain-blood partitioning of organic solutes using theoretically computed molecular descriptors and multivariate statistics. The program MolSurf was used to compute theoretical molecular descriptors related to physicochemical properties such as lipophilicity, polarity, polarizability, and hydrogen bonding. The multivariate Partial Least Squares Projections to Latent Structures (PLS) method was used to delineate the relationship between the brain-blood partitioning of organic solutes and the theoretically computed molecular descriptors. Good statistical models were derived. Properties associated with polarity and Lewis base strength had the largest impact on the blood-brain partitioning and should be kept to a minimum to promote high partitioning. The absence of atoms capable of hydrogen bonding interactions as well as high lipophilicity and the presence of polarizable surface electrons, i.e., valence electrons, were also found to promote high brain-blood partitioning. The results indicate that theoretically computed molecular MolSurf descriptors in conjunction with multivariate statistics of PLS type can be used to successfully model the brain-blood partitioning of organic solutes and hence differentiate drugs with poor partitioning from those with acceptable partitioning at an early stage of the preclinical drug-discovery process.

A quantitative structure-activity relationship study of some substance P-related peptides. A multivariate approach using PLS and variable selection.

Nine new analogues of substance P (SP) were designed using quantitative sequence-activity models based on the amino acid z-scales with PLS as the statistical method and the GOLPE procedure for variable selection. The nine SP analogues were synthesised by solid-phase peptide synthesis and tested for affinity to the NK-1 receptor from rat brain with radio receptor assay using [125I]-Bolton-Hunter substance P as labelled ligand. All of the new substance P analogues showed high affinities, with IC50 values of less than 0.8 nM. One analog, Lys-Arg-Ala-Lys-Phe-Met-Met-Phe-Phe-Gly-Leu-Let-NH2, showed a exceptional high affinity for the NK1 receptor, with IC50 = 5 pM.

Theoretical calculation and prediction of Caco-2 cell permeability using MolSurf parametrization and PLS statistics.

PURPOSE: To statistically model the permeability across Caco-2 cell monolayers using theoretically computed molecular descriptors and multivariate statistics. METHODS: Seventeen structurally diverse compounds were investigated. The program MolSurf was used to compute theoretical molecular descriptors related to physico-chemical properties such as lipophilicity, polarity, polarizability and hydrogen bonding. The multivariate Partial Least Squares Projections to Latent Structures (PLS) method was used to delineate the relationship between the permeability across Caco-2 cell monolayers and the theoretically computed molecular descriptors. RESULTS: Excellent statistical models were derived. Properties associated with hydrogen bonding had the largest impact on diffusion through the monolayers and should be kept at a minimum to promote high permeability. High lipophilicity and the presence of surface electrons, i.e. valence electrons, which are not tightly bonded to the molecule, were also found to have a favorable influence to achieve high permeability. CONCLUSIONS: The results indicate that theoretically computed molecular MolSurf descriptors in conjunction with multivariate statistics of PLS type can be used to successfully model permeability across Caco-2 cell monolayers and, thus, differentiate drugs with poor permeability from those with acceptable permeability at an early stage of the preclinical drug discovery process.

Determinants for DNA-binding site recognition by the glucocorticoid receptor.

The glucocorticoid receptor binds with high specificity to glucocorticoid response elements, discriminating them from other closely related binding sites. Three amino acids in the recognition alpha-helix of the DNA-binding domain of the receptor are primarily responsible for this specific DNA binding activity. In this study we analyze in detail how these residues determine the specific DNA binding by studying a series of mutant glucocorticoid receptor DNA-binding domains containing all combinations of glucocorticoid and estrogen receptor-specific residues at these positions. Statistical analysis of the results enables us to create models describing the association between amino acids and base pairs. Several strategies appear to be used in accomplishing discrimination between the glucocorticoid and estrogen response elements. Single residues (i.e., Val-443 in the glucocorticoid receptor and Glu-439 in the estrogen receptor) appear to form both positive contacts with specific base pairs in the cognate binding site and negative contacts in the non-cognate site. In the glucocorticoid receptor Ser-440 is pleiotropically negative for all sites tested but the negative effect is stronger for the estrogen response element thus contributing to binding site discrimination. Furthermore, combinations of amino acids appear to act synergistically, most often causing a reduction in binding to non-cognate sites.

Inhibition of [3H]paroxetine binding by various serotonin uptake inhibitors: structure-activity relationships.

Fifty-four compounds structurally related to zimeldine or alaproclate and eight reference substances were examined as inhibitors of the high affinity binding of [3H]paroxetine to rat cerebral cortical membranes as a measure of the affinity of the 5-hydroxytryptamine (5-HT) transporter. None of the compounds had an affinity as high as paroxetine (KD = 0.026 nM). The most potent compound, 3-(4-methoxyphenyl)-1-methyl-3-phenylpropylamine (2) had a 5 times lower affinity than paroxetine. Some other diphenyl-1-methyl-propylamines displayed high affinity, e.g. the 4-bromo (4) and 2-bromo (7) derivatives. The primary amine analogue of zimeldine substituted with an alpha-methyl group (19) had an affinity only slightly less than that of norzimeldine (11) but an almost 100 times higher affinity than that of the unsubstituted primary zimeldine analogue (57). These observations indicate that a methyl group on the alpha-carbon and on the nitrogen both increase the affinity for the [3H]paroxetine binding site. The structure activity relationship for the compounds to inhibit [3H]paroxetine binding was highly significantly correlated to the inhibition of 5-HT uptake in mouse brain slices (P less than 0.01) and to the inhibition of noradrenaline uptake in the same slices (P less than 0.05). QSAR analysis of the zimeldine series of compounds indicates that substitution of halogens of the 2-position of the phenyl ring is unfavourable. The cis configuration promotes higher activity than the trans configuration.

PLS-based quantitative structure-activity relationship for substituted benzamides of clebopride type. Application of experimental design in drug design.

The advantageous approach of using an experimentally designed training set as the basis for establishing a quantitative structure-activity relationship with good predictive capability is described. The training set was selected from a fractional factorial design scheme based on a principal component description of physico-chemical parameters of aromatic substituents. The derived model successfully predicts the activities of additional substituted benzamides of 6-methoxy-N-(4-piperidyl)salicylamide type. The major influence on activity of the 3-substituent is demonstrated.