Parallel Screening: A Novel Concept in Pharmacophore Modeling and Virtual Screening † Theodora M. Steindl, ‡,§ Daniela Schuster, ‡ Christian Laggner, ‡ and Thierry Langer* ,‡,§ Institute of Pharmacy, Computer Aided Molecular Design Group, University of Innsbruck, Innrain 52c, Austria, Center for Molecular Biosciences Innsbruck (CMBI), Peter-Mair-Str.1, A-6020 Innsbruck, Austria, and Inte:Ligand GmbH, Clemens Maria Hofbauer-Gasse 6, A-2344 Maria Enzersdorf, Austria Received May 18, 2006 Parallel screening comprises a novel in silico method to predict the potential biological activities of a compound by screening it with a multitude of pharmacophore models. Our aim is to provide a fast, large- scale system that allows for virtual activity profiling. In this proof of principle study, carried out with the software tools LigandScout and Catalyst, we present a model work for the application of parallel pharmacophore-based virtual screening on a set of 50 structure-based pharmacophore models built for various viral targets and 100 antiviral compounds. The latter were screened against all pharmacophore models in order to determine if their biological targets could be correctly predicted via an enrichment of corresponding pharmacophores matching these ligands. The results demonstrate that the desired enrichment, that is, successful virtual activity profiling, was achieved for approximately 90% of all input molecules. We discuss descriptors for output validation, as well as various aspects influencing the analysis of the obtained activity profiles, and the effect of the utilized search modus for screening. INTRODUCTION The generation of pharmacophore models and consequent compound/database (DB) screening to identify molecules with the desired biological effect has become a broadly applied technique in drug discovery. The classical pharma- cophore approach comprises the following steps: After definition of the target of interest, relevant data are collected from a thorough literature search, for example, structural information on the target/binding site, known active ligands, interaction patterns, and so forth. This is followed by the determination of the pharmacophoric features, that is, the functional and steric requirements for ligand binding, and consequent pharmacophore model generation. A pharma- cophore can form in the scientist’s head, on a sheet of paper, or on the computer. It not only combines and visualizes critical interactions but also serves the purpose of checking other compounds for their ability to map the required features. With the appropriate software tool, even large DBs can be screened. Of course, these models should be validated beforehand as efficiently as possible. 1-3 Parallel screening (PS), however, focuses on the ligand rather than on the target/pharmacophore model. Assuming there is a multitude of pharmacophore models representing a variety of different pharmacological targets, would it not be desirable to have a system to screen a molecule simultaneously against all of these models? One would obtain a hit list of matching pharmacophores and could link them to the biological targets, thereby enabling an in silico identification of macromolecular systems that will possibly be influenced by this ligand. That is exactly the aim that PS seeks to achieve: a compound screened against a set of high quality pharmacophore models will provide a hit list of mapping models, the so-called pharmacophoric profile. According to the targets encoded by these models, a pharmacological profile for the compound will emerge (Scheme 1). On one hand, this enables the virtual charac- terization of the biological properties of new compounds. On the other hand, the sphere of action for substances with already established activities could be enlarged, an often highly successful concept in drug development. 4,5 Of course, if the PS system includes the appropriate models, prognoses will also cover toxicity, side effects, antitargets, and meta- bolic pathways. 6-11 Another interesting aspect in PS is that the behavior of a compound in the system can point toward promiscuity and therefore be a warning signal. Further applications of this approach might be the fine-tuning of early results in high-throughput screening or ideas for the iden- tification of targets hit by natural products that are thera- peutically used because of long-time experience but not mechanistically characterized. 12-14 A system for PS requires (i) a large set of pharmacophore hypotheses including the availability of a fast and reliable tool for their automatic generation and (ii) a high-speed screening platform to test one compound against a variety of models, which should also allow for analysis, visualiza- tion, and facile interpretability of the output data. Our aim is to provide such an automatic system for fast virtual activity profiling of compounds. The number of pharmacophore hypotheses in the system is constantly growing, targeted at extensive coverage of available biologi- * To whom correspondence should be addressed. Tel.: +43 512 507 5252. Fax +43 512 507 5269. E-mail: thierry.langer@uibk.ac.at. † Parts of this study have been presented at the 9th E.U. Catalyst User Group Meeting 2006, March 23, 2006, Frankfurt, Germany, and at the 231st ACS National Meeting, March 29, 2006, Atlanta, GA, as an oral presenta- tion. ‡ University of Innsbruck and CMBI. § Inte:Ligand GmbH. 2146 J. Chem. Inf. Model. 2006, 46, 2146-2157 10.1021/ci6002043 CCC: $33.50 © 2006 American Chemical Society Published on Web 08/23/2006