Tuesday, January 11, 2005

7:55    Chairperson's Remarks
Terry R. Stouch, Ph.D., Head Computational Chemistry, Lexicon Pharmaceuticals, Inc.

Building and Using Models

8:00    Pharmacogenetic Models for Predictive Toxicology
Dr. Howard Jacob, CSO, PhysioGenix, Inc.

8:30    Computational ADME/Tox Modeling: Aiding Understanding and Enhancing Decision Making in Drug Design
Dr. Robert DeLisle, Scientist, Molecular Modeling, Pharmacopeia, Inc.
With recent estimates of drug development costs on the order of $800 million and increased pressure to reduce consumer drug costs, it is not surprising that the pharmaceutical industry is keenly interested in reducing the overall expense associated with drug development. Computational models provide a low cost, flexible evaluation of compound properties that can be implemented even prior to chemical synthesis. Here we review some of the commercially available tools to achieve this goal and, further, discuss methods developed internally to address these issues from the design stage through development and optimization of drug candidates. 

9:00    Multi-Dimensional Lead Optimization with Recursive Partitioning
Dr. Yi Han, Principal Scientist, Department: Discovery Chemistry, Hoffmann-La Roche, Ltd.
In this study, Recursive Partitioning (RP) method was used to guide lead optimization for a novel metabolic disease target that is believed to play an important role in obesity treatment. Both activity (enzyme binding IC50) and metabolic stability in human liver microsome (HLM) were modeled. RP successfully discovered rules for making lead compounds that are both active and stable.

10:00    Practices and Perspectives in Application of In Silico ADME/Tox to Drug Discovery
Dr. Terry R. Stouch, Head Computational Chemistry, Lexicon Pharmaceuticals, Inc. 
The urgent need for solutions to expensive and potentially crippling ADME/Tox issues has resulted in substantial emphasis on and expectations for in silico approaches. Apparent successes of these approaches in particular cases has fueled these expectations. However, how truly useful are these approaches when integrated into actual drug discovery and development efforts? We will discuss successes, failures, common sense, and the appropriate use of these methods.

10:30    Web-Based Cheminformatics and Property Prediction Tools at Novartis
Dr. Peter Ertl , Head of Cheminformatics, Novartis Institute for BioMedical Research
Web-based molecular processing and property prediction tools installed on corporate intranets bring easy to use cheminformatics and molecular modelling capabilities directly to the desks of synthetic chemists, considerably improving efficiency of the drug design and development process. User friendly tools using standard web browser as interface allow users access to a broad range of expert molecular processing tools and techniques, without the need for extensive expertise in their use. In the presentation the Novartis web-based cheminformatics system consisting of more than 20 modules, supporting a broad range of various molecular processing tasks will be presented. Focus will be given on prediction of molecular ADME properties, including water solubility, intestinal absorption or plasma protein binding. Also a sophisticated toxicity alerting system will be described. 

11:00    Panel Discussion: The Challenge of Building Models that Work

11:45    Technology Workshop (Sponsorship Available)

12:15    Lunch on your own (Sponsorship Available)

1:40    Chairperson's Opening Remarks
Dr. Craig M. Zwickl, Senior Toxicologist, Eli Lilly and Company

Building Predictive Toxicology Models

1:45    Strategies for Building Predictive Models in Toxicology
Dr. Paul Blower, Chief Scientific Officer, LeadScope, Inc.
Building predictive models derived from experimental toxicology data is challenging. Structure-based clustering often reveals an irregular landscape, both in terms of the compound classes represented and the distribution of active compounds across structural classes. Even within structurally homogeneous classes, the ratio of actives to inactives is often out of balance, or the range of response values too narrow, or classes are too small to derive accurate models. This presentation will survey problems of building predictive models from public toxicology data and describe our approaches to data assessment, identifying modelable structural neighborhoods, model building and validation, and defining domains of applicability. In collaboration with Kevin Cross, Glenn Myatt, and Chihae Yang.

2:15    Automated Building of Machine-Learning Models to Predict Toxicity
Dr. Boyd Steere, Senior Scientist, Department of Life Sciences, Simulations Plus, Inc.
I will discuss how researchers who have no formal training in machine learning methods can use data from their own studies to build predictive toxicity models automatically. My talk covers the difference between local and global models, introduces neural networks and support vector machines, and the method that the algorithm uses to build the models. The talking points are illustrated through case-study examples.

2:45    Refreshment Break, Poster and Exhibit Viewing

Data Collection and Interpretation

3:15    Predicting Side Effects of New Molecular Entities: The BioPrint® Approach
Dr. Jacques Migeon, Senior Scientist, Department: BioPrint, Cerep, Inc.
The BioPrint® database is a large and unique collection in vitro (over 150 assays) and in vivo data for over 2000 drugs and drug-like molecules. Statistically robust in vitro-in vivo associations relating risk of specific side effects with specific pharmacological activities were developed.

3:45    Reading the Transcriptome, from the Perspective of a Pathologist Interested in Dynamics
Dr. Kevin Thomas Morgan, Scientist, Sanofi - Aventis 
Transient states of entire transcriptomes are becoming available for integration with other endpoints during the assessment of disease states. The application of such studies to Toxicology has become known as Toxicogenomics. Interpreting these temporally sparse data sets is challenging, exciting, and above all, educational. Interesting mechanistic pathobiology is often revealed by gene expression changes exhibited in such things as poisoned livers and infected lungs. When combined with a range of Bioinformatics approaches, the 'one-gene-at-a-time' examination of transcriptome data sets is highly recommended to toxicologists, pathologists, and other biologists involved in human safety assessment of chemicals, including drugs. Such studies should ideally be combined with an understanding of underlying dynamics, which can be greatly assisted by the development of mathematical models of the behavior of critical biological components of the system of interest.

4:15    Strategies to Enable Data-Sharing for Predictive Toxicology Model Building
Dr. Craig M. Zwickl, Senior Toxicologist, Eli Lilly and Company 
The thoughtful application of predictive methods, including in silico modeling, to toxicology will undoubtedly be an essential element of any new approach designed to reduce the cost and increase the probability of technical success associated with identifying and developing new drugs. Although mathematical methods for correlating biological effects with chemical descriptors are well-developed, in practice, our ability to build predictive in silico models to support early discovery efforts continues to suffer from a relative dearth of data covering a suitably diverse chemical space. Recent initiatives to create electronic data standards and controlled vocabularies, although intended to enable electronic data submission to regulatory agencies, also lays a foundation that can be leveraged to enable data sharing within the industry. Before this can be practically realized however, a key challenge will be to determine how to share useful data without jeopardizing intellectual property position.

4:45    Facilitated Networking Roundtables
Discussion Topics:  • Data Quality & Data Sharing • Bioavailability and Absorption • "Lost in Translation" - Defining "Building a Predictive Model"

5:45    End of Predictive ADME conference