Lothar Terfloth

QSAR Modeling: Where have you been? Where are you going to?

Quantitative structure-activity relationship modeling is one of the major computational tools employed in medicinal chemistry. However, throughout its entire history it has drawn both praise and criticism concerning its reliability, limitations, successes, and failures. In this paper, we discuss (i) the development and evolution of QSAR; (ii) the current trends, unsolved problems, and pressing challenges; and (iii) several novel and emerging applications of QSAR modeling. Throughout this discussion, we provide guidelines for QSAR development, validation, and application, which are summarized in best practices for building rigorously validated and externally predictive QSAR models. We hope that this Perspective will help communications between computational and experimental chemists toward collaborative development and use of QSAR models. We also believe that the guidelines presented here will help journal editors and reviewers apply more stringent scientific standards to manuscripts reporting new QSAR studies, as well as encourage the use of high quality, validated QSARs for regulatory decision making.

Identification of Novel Functional Inhibitors of Acid Sphingomyelinase

We describe a hitherto unknown feature for 27 small drug-like molecules, namely functional inhibition of acid sphingomyelinase (ASM). These entities named FIASMAs (Functional Inhibitors of Acid SphingoMyelinAse), therefore, can be potentially used to treat diseases associated with enhanced activity of ASM, such as Alzheimers disease, major depression, radiation- and chemotherapy-induced apoptosis and endotoxic shock syndrome. Residual activity of ASM measured in the presence of 10 µM drug concentration shows a bimodal distribution; thus the tested drugs can be classified into two groups with lower and higher inhibitory activity. All FIASMAs share distinct physicochemical properties in showing lipophilic and weakly basic properties. Hierarchical clustering of Tanimoto coefficients revealed that FIASMAs occur among drugs of various chemical scaffolds. Moreover, FIASMAs more frequently violate Lipinskis Rule-of-Five than compounds without effect on ASM. Inhibition of ASM appears to be associated with good permeability across the blood-brain barrier. In the present investigation, we developed a novel structure-property-activity relationship by using a random forest-based binary classification learner. Virtual screening revealed that only six out of 768 (0.78%) compounds of natural products functionally inhibit ASM, whereas this inhibitory activity occurs in 135 out of 2028 (6.66%) drugs licensed for medical use in humans.

Ligand-based models for the isoform specificity of cytochrome P450 3A4, 2D6, and 2C9 substrates

A data set of 379 drugs and drug analogs that are metabolized by human cytochrome P450 (CYP) isoforms 3A4, 2D6, and 2C9, respectively, was studied. A series of descriptor sets directly calculable from the constitution of these drugs was systematically investigated as to their power into classifying a compound into the CYP isoform that metabolizes it. In a four-step build-up process eventually 303 different descriptor components were investigated for 146 compounds of a training set by various model building methods, such as multinomal logistic regression, decision tree, or support vector machine (SVM). Automatic variable selection algorithms were used in order to decrease the number of descriptors. A comprehensive scheme of cross-validation (CV) experiments was applied to assess the robustness and reliability of the four models developed. In addition, the predictive power of the four models presented in this paper was inspected by predicting an external validation data set with 233 compounds. The best model has a leave-one-out (LOO) cross-validated predictivity of 89% and gives 83% correct predictions for the external validation data set. For our favored model we showed the strong influence on the predictivity of the way a data set is split into a training and test data set.

Neural networks and genetic algorithms in drug design

Abstract Neural networks and genetic algorithms are versatile methods for a variety of tasks in rational drug design, including analysis of structure–activity data, establishment of quantitative structure–activity relationships (QSAR), gene prediction, locating protein-coding regions in DNA sequences, 3D structure alignment, pharmacophore perception, docking of ligands to receptors, automated generation of small organic compounds, and the design of combinatorial libraries. Here, we give a brief overview of these applications of neural networks and genetic algorithms in drug design, and provide an insight into the underlying principles of such methods.

New Publicly Available Chemical Query Language, CSRML, To Support Chemotype Representations for Application to Data Mining and Modeling

Chemotypes are a new approach for representing molecules, chemical substructures and patterns, reaction rules, and reactions. Chemotypes are capable of integrating types of information beyond what is possible using current representation methods (e.g., SMARTS patterns) or reaction transformations (e.g., SMIRKS, reaction SMILES). Chemotypes are expressed in the XML-based Chemical Subgraphs and Reactions Markup Language (CSRML), and can be encoded not only with connectivity and topology but also with properties of atoms, bonds, electronic systems, or molecules. CSRML has been developed in parallel with a public set of chemotypes, i.e., the ToxPrint chemotypes, which are designed to provide excellent coverage of environmental, regulatory, and commercial-use chemical space, as well as to represent chemical patterns and properties especially relevant to various toxicity concerns. A software application, ChemoTyper has also been developed and made publicly available in order to enable chemotype searching and fingerprinting against a target structure set. The public ChemoTyper houses the ToxPrint chemotype CSRML dictionary, as well as reference implementation so that the query specifications may be adopted by other chemical structure knowledge systems. The full specifications of the XML-based CSRML standard used to express chemotypes are publicly available to facilitate and encourage the exchange of structural knowledge.

Comparison of Multilabel and Single-Label Classification Applied to the Prediction of the Isoform Specificity of Cytochrome P450 Substrates

Each drug can potentially be metabolized by different CYP450 isoforms. In the development of new drugs, the prediction of the metabolic fate is important to prevent drug-drug interactions. In the present study, a collection of 580 CYP450 substrates is deeply analyzed by applying multi- and single-label classification strategies, after the computation and selection of suitable molecular descriptors. Cross-training with support vector machine, multilabel k-nearest-neighbor and counterpropagation neural network modeling methods were used in the multilabel approach, which allows one to classify the compounds simultaneously in multiple classes. In the single-label models, automatic variable selection was combined with various cross-validation experiments and modeling techniques. Moreover, the reliability of both multi- and single-label models was assessed by the prediction of an external test set. Finally, the predicted results of the best models were compared to show that, even if the models present similar performances, the multilabel approach more coherently reflects the real metabolism information.

9 – ELECTRONIC SCREENING: LEAD FINDING FROM DATABASE MINING

Fayyad defined and described the term “data mining” as the “nontrivial extraction of implicit, previously unknown, and potentially useful information from data, or the search for relationships and global patterns that exist in databases.” To extract information from huge quantities of data and to gain knowledge from this information, analysis and exploration have to be performed by automatic or semi-automatic methods. The data mining process can be divided into the following steps: selection, preprocessing, transformation, interpretation, and evaluation. In particular, for the final two steps, the visual representation of data plays a pivotal role. The task of finding a lead by database mining requires the analysis of the relationships between the structure of potential new drugs and their biological activity. Due to the amount of data to be processed, it is advisable to use a hierarchical representation of the chemical structures starting from 1D fingerprints, going further to topological descriptors, such as 2D autocorrelation, and finally, considering 3D structures and molecular surface properties. A structure can be searched for in a database by string matching, as long as each compound has a unique Wiswesser line notation (WLN) or a unique simplified molecular input line system (SMILES) string. The SMILES arbitrary target specification (SMARTS) is based on the SMILES notation and is used to encode a query for substructure searches.

Exploring potency and selectivity receptor antagonist profiles using a multilabel classification approach: the human adenosine receptors as a key study.

Nowadays, in medicinal chemistry adenosine receptors represent some of the most studied targets, and there is growing interest on the different adenosine receptor (AR) subtypes. The AR subtypes selectivity is highly desired in the development of potent ligands to achieve the therapeutic success. So far, very few ligand-based strategies have been investigated to predict the receptor subtypes selectivity. In the present study, we have carried out a novel application of the multilabel classification approach by combining our recently reported autocorrelated molecular descriptors encoding for the molecular electrostatic potential (autoMEP) with support vector machines (SVMs). Three valuable models, based on decreasing thresholds of potency, have been generated as in series quantitative sieves for the simultaneous prediction of the hA(1)R, hA(2A)R, hA(2B)R, and hA(3)R subtypes potency profile and selectivity of a large collection, more than 500, of known inverse agonists such as xanthine, pyrazolo-triazolo-pyrimidine, and triazolo-pyrimidine analogues. The robustness and reliability of our multilabel classification models were assessed by predicting an internal test set. Finally, we have applied our strategy to 13 newly synthesized pyrazolo-triazolo-pyrimidine derivatives inferring their full adenosine receptor potency spectrum and hAR subtypes selectivity profile.

Applications of integrated data mining methods to exploring natural product space for acetylcholinesterase inhibitors.

Nature, especially the plant kingdom, is a rich source for novel bioactive compounds that can be used as lead compounds for drug development. In order to exploit this resource, the two neural network-based virtual screening techniques novelty detection with self-organizing maps (SOMs) and counterpropagation neural network were evaluated as tools for efficient lead structure discovery. As application scenario, significant descriptors for acetylcholinesterase (AChE) inhibitors were determined and used for model building, theoretical model validation, and virtual screening. Top-ranked virtual hits from both approaches were docked into the AChE binding site to approve the initial hits. Finally, in vitro testing of selected compounds led to the identification of forsythoside A and (+)-sesamolin as novel AChE inhibitors.

Neural Networks as Valuable Tools To Differentiate between Sesquiterpene Lactones’ Inhibitory Activity on Serotonin Release and on NF-κB

Sesquiterpene lactones are the active components of a variety of medicinal plants from the Asteraceae family. They possess biological activities such as the inhibition of NF-kappaB and the release inhibition of the vasoactive serotonin. On the basis of a data set of 54 SLs, we report the development of a quantitative model for the prediction of serotonin release inhibition. Comparing this model with a previous investigation of the target NF-kappaB, structural features necessary for specific compounds could be acquired. Atomic properties encoded by radial distribution function and molecular surface potentials encoded by autocorrelation were used as descriptors. Whereas some descriptors describe the structural requirements for both activities, other descriptors can be used to decide whether an SL is more active to NF-kappaB or to serotonin release. Again, counter propagation neural networks proved to be a valuable tool to establish structure-activity relationships that are necessary for the search for and optimization of lead structures.