Bernd Beck

QM/NN QSPR models with error estimation: vapor pressure and logP

QSPR models for logP and vapor pressures of organic compounds based on neural net interpretation of descriptors derived from quantum mechanical (semiempirical MO; AM1) calculations are presented. The models are cross-validated by dividing the compound set into several equal portions and training several individual multilayer feedforward neural nets (trained by the back-propagation of errors algorithm), each with a different portion as test set. The results of these nets are combined to give a mean predicted property value and a standard deviation. The performance of two models, for logP and the vapor pressure at room temperature, is analyzed, and the reliability of the predictions is tested.

A support vector machine approach to classify human cytochrome P450 3A4 inhibitors

SummaryThe cytochrome P450 (CYP) enzyme superfamily plays a major role in the metabolism of commercially available drugs. Inhibition of these enzymes by a drug may result in a plasma level increase of another drug, thus leading to unwanted drug–drug interactions when two or more drugs are coadministered. Therefore, fast and reliable in silico methods predicting CYP inhibition from calculated molecular properties are an important tool which can be applied to assess both already synthesized as well as virtual compounds. We have studied the performance of support vector machines (SVMs) to classify compounds according to their potency to inhibit CYP3A4. The data set for model generation consists of more than 1300 structural diverse drug-like research molecules which were divided into training and test sets. The predictive power of SVMs crucially depends on a careful selection of parameters specifying the kernel function and the penalty for misclassifications. In this study we have investigated a procedure to identify a valid set of SVM parameters which is based on a sampling of the parameter space on a regular grid. From this set of parameters, either single SVMs or SVM committees were trained to distinguish between strong and weak inhibitors or to achieve a more realistic three-class assignment, with one class representing medium inhibitors. This workflow was studied for several kernel functions and descriptor sets. All SVM models performed significantly better than PLS-DA models which were generated from the corresponding descriptor sets. As a very promising result, simple two-dimensional (2D) descriptors yield a three-class model which correctly classifies more than 70% of the test set. Our work illustrates that SVMs used in combination with simple 2D descriptors provide a very effective and reliable tool which allows a fast assessment of CYP3A4 inhibition potency in an early in silico filtering process.

A composite model for HERG blockade

hERG blockade is one of the major toxicological problems in lead structure optimization. Reliable ligand‐based in silico models for predicting hERG blockade therefore have considerable potential for saving time and money, as patch‐clamp measurements are very expensive and no crystal structures of the hERG‐encoded channel are available. Herein we present a predictive QSAR model for hERG blockade that differentiates between specific and nonspecific binding. Specific binders are identified by preliminary pharmacophore scanning. In addition to descriptor‐based models for the compounds selected as hitting one of two different pharmacophores, we also use a model for nonspecific binding that reproduces blocking properties of molecules that do not fit either of the two pharmacophores. PLS and SVR models based on interpretable quantum mechanically derived descriptors on a literature dataset of 113 molecules reach overall R2 values between 0.60 and 0.70 for independent validation sets and R2 values between 0.39 and 0.76 after partitioning according to the pharmacophore search for the test sets. Our findings suggest that hERG blockade may occur through different types of binding, so that several different models may be necessary to assess hERG toxicity.

A quantum mechanical/neural net model for boiling points with error estimation.

We present QSPR models for normal boiling points employing a neural network approach and descriptors calculated using semiempirical MO theory (AM1 and PM3). These models are based on a data set of 6000 compounds with widely varying functionality and should therefore be applicable to a diverse range of systems. We include cross-validation by simultaneously training 10 different networks, each with different training and test sets. The predicted boiling point is given by the mean of the 10 results, and the individual error of each compound is related to the standard deviation of these predictions. For our best model we find that the standard deviation of the training error is 16.5 K for 6000 compounds and the correlation coefficient (R2) between our prediction and experiment is 0.96. We also examine the effect of different conformations and tautomerism on our calculated results. Large deviations between our predictions and experiment can generally be explained by experimental errors or problems with the semiempirical methods.

The natural atomic orbital point charge model for PM3: multipole moments and molecular electrostatic potentials

The natural atomic orbital/point (NAO‐PC) model originally developed to calculate molecular electrostatic potentials (MEPs) and multiple moments based on the AM1 wave function has been extended to PM3. As for AM1, NAO‐PC/PM3 reproduces dipole moments calculated by the standard PM3 method very well. There is also a surprisingly good correlation between experimental and calculated quadrupole moments. The MEPs calculated using PM3/NAO‐PC are found to be in better agreement with those given by RHF/6‐31G* than those obtained from the PM3 wave function using Coulson charges. On the other hand, the NAO‐PC model is often slightly worse then the method implemented in MOPAC‐ESP. The MEPs calculated using our model based on the PM3 wave function are often in better agreement with those given by RHF/6‐31G* than those obtain with AM1.

VESPA: A new, fast approach to electrostatic potential‐derived atomic charges from semiempirical methods

An improved semiempirical method for computing electrostatic potential‐derived atomic charges is described. It includes a very fast algorithm for the generation of the grid points around the molecule and the calculation of the electrostatic potential at these points. The dependency of the atomic point charges obtained on the number of grid points used in the fitting procedure is examined. For “buried” atoms a high density grid is necessary. It is possible to obtain 6–31G*‐quality atom‐centered point charges, even for phosphorus compounds, using AM1 or PM3. This approach can therefore be recommended for general use in QSAR or molecular mechanics for any organic and bioorganic system up to about 200 atoms.

A Temperature-Dependent Quantum Mechanical/Neural Net Model for Vapor Pressure

We present a temperature-dependent model for vapor pressure based on a feed-forward neural net and descriptors calculated using AM1 semiempirical MO-theory. This model is based on a set of 7681 measurements at various temperatures performed on 2349 molecules. We employ a 10-fold cross-validation scheme that allows us to estimate errors for individual predictions. For the training set we find a standard deviation of the error s = 0.322 and a correlation coefficient (R(2)) of 0.976. The corresponding values for the validation set are s = 0.326 and R(2) = 0.976. We thoroughly investigate the temperature-dependence of our predictions to ensure that our model behaves in a physically reasonable manner. As a further test of temperature-dependence, we also examine the accuracy of our vapor pressure model in predicting the related physical properties, the boiling point, and the enthalpy of vaporization.

A Consistent Dataset of Kinetic Solubilities for Early-Phase Drug Discovery

Herein, we describe a new dataset of kinetic aqueous solubilities determined by nephelometry for 711 druglike compounds. The solubilities are reported in twelve classes ranging from <2 μg mL−1 to >250 μg mL−1. The measurements were designed to provide the appropriate data for applications in the early phases of drug discovery. Three class classification models (insoluble, moderately soluble, soluble) were built using the random forest algorithm and their performance for this dataset was analyzed.

Enhanced 3D-Databases: A Fully Electrostatic Database of AM1-Optimized Structures

In a feasibility study, a single conformation 3D version of the Maybridge database (53 471 compounds) has been produced using geometries optimized with AM1 semiempirical MO-theory. The database entries include full electrostatic information within the NAO-PC model and can be used to generate spectroscopic and physical properties using established QSPR models. The data were generated from the original database using custom cleanup software to remove database inconsistencies and, for instance, to isolate the “interesting” ion of ion pairs, 2D to 3D conversion using CORINA and subsequent geometry optimization using VAMP. The complete geometry optimization run was carried out in less than 15 h elapsed time on a Silicon Graphics Origin 2000 with 126 processors. The total failure rate for the structure cleanup, 2D to 3D conversion, and geometry optimization steps was around 1%.

Insolubility classification with accurate prediction probabilities using a MetaClassifier.

Insolubility is a crucial issue in drug design because insoluble compounds are often measured to be inactive although they might be active if they were soluble. We provide and analyze various insolubility classification models based on a recently published data set and compounds measured in-house at Boehringer-Ingelheim. The 2D descriptor sets from pharmacophore fingerprints and MOE and the 3D descriptor sets from ParaSurf and VolSurf were examined in conjunction with support vector machines, Bayesian regularized neural networks, and random forests. We introduce a classifier-fusion strategy, called metaclassifier, which improves upon the best single prediction and at the same time avoids descriptor selection, a potential source of overfitting. The metaclassifier strategy is compared to the simpler fusion strategies of maximum vote and highest probability picking. A prediction accuracy of 72.6% on a three class model is achieved with the metaclassifier, with nearly perfect separation of soluble and insoluble compounds and prediction as good as our calculated maximum possible agreement with experiment.