David Lloyd, Trinity College Dublin, chaired the following QSAR session at the eCheminfo 2004 Web conference (http://echeminfo.com/), 8-19 November 2004:
Paul Blower, Leadscope, Correlating gene expression with classes of cytotoxic agents: identifying substrates and inhibitors of mdr1
Genomic studies are producing large databases of molecular information on cancers and other cell and tissue types. In order to utilize this data for drug discovery, we have developed a general analytical method for discovering relationships between compound classes and potential molecular targets. The procedure will be illustrated using the NCI-60 cell lines and microarray-based gene expression patterns. This provides a basis for associating groups of potentially novel compound classes with relevant gene families, followed by experimental validation. Our procedure involves the following steps: 1) Select an initial gene from the NCI60 gene expression database (in this case focused on genes involved in membrane transport); 2) use correlations between gene expression and drug potency to select a compound class; 3) Develop structure activity relationships within the compound class; 4) Query a genome-wide database to expand the molecular hypothesis on mechanism of cytotoxcity; 5) Validate the hypothesis experimentally; 6) Design new compounds for testing based on new molecular insights to avoid chemoresistance or enhance targeted chemosensitivity.
Michael B. Bolger, Simulations Plus, Inc., ADMET Predictions by Linking QSPR and Simulation
Oral bioavailability and biological activity of drugs can be broken down into seven components that are important considerations for drug development. The components are: Liberation from formulated product (dissolution, particle size, diffusivity, pKa, pH, food), Absorption from the lumen (permeability, lipophilicity, gastric emptying, intestinal transit, active transport, pH, pKa, protein binding), intestinal and hepatic first-pass Metabolism, Distribution into tissues, and subsequent Excretion, Response and Toxicity (LADMERT). The three computational tools that are employed for LADMERT predictions are: (1) statistical methods of quantitative molecular property relationships (QMPR) for biopharmaceutical property estimation, (2) computational chemistry methods for calculation of metabolism and enzyme-substrate interactions, and (3) mechanism-based physiological simulation methods for prediction of fraction absorbed and bioavailability. This seminar will focus on in silico estimation of log P, pKa, permeability, solubility, diffusivity, volume of distribution, blood-brain barrier penetration and protein binding. These properties have been shown to be the core properties required in computational biopharmaceutics. In addition, we will discuss the development of toxicity models based on a database of estrogen receptor ligands and on the maximum recommended therapeutic dose (MRTD). Finally, we will discuss a simulation of the gastrointestinal tract using the Advanced Compartmental Absorption and Transit (ACAT) model. This model has been applied to the prediction of fraction absorbed and the impact of physiological and biochemical factors on bioavailability. These in silico approaches have the potential to save valuable resources in the drug discovery, development, and regulatory processes.
Curt Breneman, Rensselaer Polytechnic Institute (NY), Graphical and automated analysis of the chemical information within hybrid shape/property descriptors in kernel-based models with feature selection
The field of predictive molecular property modeling spans a range of important subdisciplines within cheminformatics, and often results in the development of models that are difficult to interpret using traditional chemical intuition. Attempts to simplify or limit the number of descriptors used in a particular model result in tools that are more intuitively satisfying, but typically less capable of making validated predictions of molecular behavior to a useful level of accuracy. To address this deficiency, new tools for extracting chemical intuition from complex non-linear support vector machine (SVM) and kernel PLS (KPLS) models are to be presented. This toolkit includes graphical means for identifying key portions of molecules and pharmacophoric relationships that are represented within the model by combinations of otherwise non-intuitive descriptors. Through the development of these tools, the range of modeling techniques that can yield high-quality predictions as well as interpretable chemical knowledge will expand, and the applicability of these techniques will grow to incorporate a more diverse group of users – one that includes laboratory bench chemists.
One particular class of descriptors that has been found to contain useful chemical information are shape/property hybrids produced by the PEST program. When coupled with SVM or KPLS modeling methods, these alignment-free descriptors have been shown to outperform 3D QSAR methods such as CoMFA, without the need for cumbersome alignment rules. The presentation will illustrate some of these examples, and also demonstrate the use of graphical tools for extracting chemical meaning from descriptor patterns identified as important in the modeling process.
The recorded seminars and discussion are available through the eCheminfo site.
Barry Hardy
Douglas Connect
www.douglasconnect.com
Cheminformatics & Chemical Modelling in Drug Discovery: http://echeminfo.com/
Recent Comments