Nanoparticle design characterized by in silico preparation parameter prediction using ensemble models.

Nanoparticles (NPs) are now widely applied in new drug delivery and targeting strategies. A predictive tool for the carrier design would allow for reducing the number of experiments to determine the optimal formulation. Here we investigated the performance of two different statistical approaches to predicting NP properties. NPs were prepared by an oil/water emulsification method using Eudragit RS or poly(lactide-co-glycolide) (PLGA) as matrix polymer and dichloromethane (DCM) or ethyl acetate (EA) as organic solvent while ibuprofen was entrapped as model drug. Statistical analysis on the impact of the various formulations and process on the particle properties was performed using response surface methodology, and linear and nonlinear ensemble models. Particle size diminished with EA and the use of Eudragit RS (RS + EA: 50-100 nm; RS + DCM: 200-400 nm; PLGA + EA: 100-800 nm; PLGA + DCM: 200-1000 nm). Zeta potential was around zero for PLGA and positive with Eudragit RS. Encapsulation rates were generally higher than 80% with the tendency to increase with larger particles. Values predicted using response surface modeling or nonlinear ensemble models exhibited a high correlation with experimental values. Especially the more recent nonlinear ensemble models may be a valuable approach to facilitate and speed up the otherwise very time-consuming process of NP design for drug delivery.

A fully computational model for predicting percutaneous drug absorption.

The prediction of transdermal absorption for arbitrary penetrant structures has several important applications in the pharmaceutical industry. We propose a new data-driven, predictive model for skin permeability coefficients k(p) based on an ensemble model using k-nearest-neighbor models and ridge regression. The model was trained and validated with a newly assembled data set containing experimental data and structures for 110 compounds. On the basis of three purely computational descriptors (molecular weight, calculated octanol/water partition coefficient, and solvation free energy), we have developed a model allowing for the reliable, purely computational prediction of skin permeability coefficients. The model is both accurate and robust, as we showed in an extensive validation (correlation coefficient for leave-one-out cross validation: Q = 0.948, mean standard error: 0.2 for log k(p)).

Automatic generation of complementary descriptors with molecular graph networks.

We describe a method for the automatic generation of weakly correlated descriptors for molecular data sets. The method can be regarded as a statistical learning procedure that turns the molecular graph, representing the 2D formula of the compound, into an adaptive whole molecule composite descriptor. By translating the molecular graph structure into a dynamical system, the algorithm can compute an output value that is highly sensitive to the molecular topology. This system can be trained by gradient descent techniques, which rely on the efficient calculation of the gradient by back-propagation. We present computational experiments concerning the classification of the Developmental Therapeutics Program AIDS antiviral screen data set on which the performance of the method compares with that of approaches based on substructure comparison.

POEM: Parameter Optimization using Ensemble Methods: application to target specific scoring functions.

In computational biology processes such as docking, binding, and folding are often described by simplified, empirical models. These models are fitted to physical properties of the process by adjustable parameters. An appropriate choice of these parameters is crucial for the quality of the models. Locating the best choices for the parameters is often is a difficult task, depending on the complexity of the model. We describe a new method and program, POEM (Parameter Optimization using Ensemble Methods), for this task. In POEM we combine the DOE (Design Of Experiment) procedure with ensembles of different regression methods. We apply the method to the optimization of target specific scoring functions in molecular docking. The method consists of an iterative procedure that uses alternate evaluation and prediction steps. During each cycle of optimization we fit an approximate function to a defined loss function landscape and improve the quality of this fit from cycle to cycle by constantly augmenting our data set. As test applications we fitted the FlexX and Screenscore scoring functions to the kinase and ATPase protein classes. The results are promising: Starting from random parameters we are able to locate parameter sets which show superior performance compared to the original values. The POEM approach converges quickly and the approximated loss function landscapes are smooth, thus making the approach a suitable method for optimizations on rugged landscapes.

Thrombin inhibitors identified by computer-assisted multiparameter design.

Here, we present a series of thrombin inhibitors that were generated by using powerful computer-assisted multiparameter optimization process. The process was organized in design cycles, starting with a set of randomly chosen molecules. Each cycle combined combinatorial synthesis, multiparameter characterization of compounds in a variety of bioassays, and algorithmic processing of the data to devise a set of compounds to be synthesized in the next cycle. The identified lead compounds exhibited thrombin inhibitory constants in the lower nanomolar range. They are by far the most selective synthetic thrombin inhibitors, with selectivities of >100,000-fold toward other proteases such as Factor Xa, Factor XIIa, urokinase, plasmin, and Plasma kallikrein. Furthermore, these compounds exhibit a favorable profile, comprising nontoxicity, high metabolic stability, low serum protein binding, good solubility, high anticoagulant activity, and a slow and exclusively renal elimination from the circulation in a rat model. Finally, x-ray crystallographic analysis of a thrombin-inhibitor complex revealed a binding mode with a neutral moiety in the S1 pocket of thrombin.

Ensemble methods for classification in cheminformatics.

We describe the application of ensemble methods to binary classification problems on two pharmaceutical compound data sets. Several variants of single and ensembles models of k-nearest neighbors classifiers, support vector machines (SVMs), and single ridge regression models are compared. All methods exhibit robust classification even when more features are given than observations. On two data sets dealing with specific properties of drug-like substances (cytochrome P450 inhibition and "Frequent Hitters", i.e., unspecific protein inhibition), we achieve classification rates above 90%. We are able to reduce the cross-validated misclassification rate for the Frequent Hitters problem by a factor of 2 compared to previous results obtained for the same data set with different modeling techniques.