Cheminformatic Tools for Medicinal Chemists

IntroductionCheminformatics can be broadly described as any attempttousechemicalinformationtoinfertherelationshipsbetweenor attributes of chemical structures. From a drug discoveryperspective, cheminformatic principles can be applied fromthe earliest stages of lead discovery (e.g., chemical similarityand library design) to lead optimization (e.g., QSAR studies)through to preclinical and clinical development (e.g., predic-tive toxicology). The popularity of cheminformatics and itsuse in academia and the pharmaceutical industry can beappreciated from the fact that at least five scientific journalsexist almost exclusively dedicated to the field (The Journal ofCheminformatics, The Journal of Chemical Information andModeling, The Journal of Computer-Aided Molecular Design,Molecular Bioinformatics,andQSAR and CombinatorialScience), and more than 15000 scientific journal articles havebeen published during just the last 5 years that describecheminformatic research. This intense interest in cheminfor-matics stems from the promise that, if underlying relation-ships between a given chemical structure and a host ofbiological end points exist and can be elucidated, drug dis-covery timelines can be significantly reduced. Given thepressure on the pharmaceutical industry to increase produc-tivity while decreasing costs, prior knowledge of which mole-cules have the highest probability of success (or at leastknowing which molecules are likely to fail) is worthy ofvigorous pursuit.Over the past decade there have been several significantadvancements in our understanding and application of che-minformatic principles. Approaches to measuring and com-paring chemical information have become both moresophisticated and accessible. For example, two of the mostpowerfulchemicalsimilaritymeasures(two-dimensional(2D)extended connectivity fingerprints and three-dimensional(3D) shape and electrostatic overlays) are available in user-friendly software packages from Scitegic (Accelrys) andOpeneye Scientific Software. Multiple methods for under-standing and predictingbioactivity have proven their robust-ness,includingpartialleast-squares(PLS),geneticalgorithms,Bayesian analyses, and Random Forest analyses. Our under-standing of molecular features or properties associated withcertain pharmacological end points has also dramaticallyincreased. For example, it has been widely recognized thatcertain structural features can be associated with toxicity,while other molecular properties (such as ClogP, molecularweight, and polar surface area) can be associated with oralbioavailability