Immediately Following Cambridge Healthtech Institute's 1st Annual Predictive ADME Conference

Wednesday, November 19, 2003

8:00 Registration, Light Continental Breakfast

 

Joint Session ADME/Tox Models: The Interface 8:30 Comments by Session Chairperson Dr. Julie Penzotti, Director, Computational Chemistry, Rational Discovery, LLC 8:45 Applying Artificial Intelligence to ADMET Predictions: A Collaborative Approach Towards Improved Models Dr. David E. Clark, Argenta Discovery Ltd. The next generation of predictive ADMET models will be driven by: Improvements in molecular descriptors and statistical modeling methods Improvements in the quality and quantity of data available for model building This talk will describe a collaborative research project between Amedis Pharmaceuticals and Argenta Discovery aiming to address both of these issues: Amedis Pharmaceuticals has developed novel, proprietary molecular descriptors and modelling technologies, based on genetic programming and advanced statistical techniques Argenta Discovery is applying its in vitro ADMET expertise to generate high quality data, e.g., for inhibition of various cytochrome P450s Details of this collaboration and its progress towards improved models will be presented. 9:15 Predictive Modeling Techniques Being Used by Companies Dr. Ravi Mallela, Recommind Inc. Predictive modeling is still in its infancy. A variety of methods has been developed in an attempt to answer the challenge. This talk will provide real-life case studies that focus on the currently used methods, the advantages and disadvantages, and the conclusions we can draw from them. 9:45 Integrating ADME/Tox Prediction and Informatics Dr. Gregory Banik, General Manager, Informatics Division, Bio-Rad Laboratories, Inc. ADME/Tox prediction is becoming an increasingly important tool in the evaluation and prioritization of compounds early in the drug discovery process. A system will be described that facilitates the deployment of multiple simultaneous prediction approaches to maximize prediction accuracy. The applicability of this to high-throughput evaluation will be discussed, as will the informatics issues relevant to in silico prediction, along with possible solutions. 10:15 Poster and Exhibit Viewing, Refreshment Break 10:45 Whole Body Autoradiography Answers Pivotal Questions for Early ADME and Toxicology Dr. Eric Solon, Director of Autoradiography and Animal Resources, Quest Pharmaceutical Services Whole Body Autoradiography provides tissue distribution information and can help to identify target organs and early information regarding ADME properties. Known drug interactions can be employed in the study design and answer more specific questions. Several examples of how QWBA has been used in drug discovery for ADME and Tox support will be presented. 11:15 An Ensemble Approach to Building Predictive ADME/Tox Models for Decision Support Dr. Julie Penzotti, Director, Computational Chemistry, Rational Discovery, LLC The development of predictive models for ADME/Tox properties provides decision-support tools that can be used in library design, lead optimization and the selection of drug candidates for development. Because multiple (possibly unknown) mechanisms can lead to the same toxicity endpoint or ADME property, algorithms for their prediction must be capable of handling multiple modes of action. Our modeling methods for ADME/Tox combine computational chemistry with an ensemble approach to machine learning, thereby selecting useful molecular descriptors and multiple predictive models to produce a single composite model that generally exhibits greater predictive power and generality. 11:45 Panel Discussion with all Morning Speakers 12:15 Conclusion of Predictive ADME Conference

ADME/Tox Models: The Interface
Joint Session

8:30 Comments by Session Chairperson
Dr. Julie Penzotti, Director, Computational Chemistry, Rational Discovery, LLC

8:45 Applying Artificial Intelligence to ADMET Predictions: A Collaborative Approach Towards Improved Models
Dr. David E. Clark, Argenta Discovery Ltd.
The next generation of predictive ADMET models will be driven by:

• Improvements in molecular descriptors and statistical modeling methods
• Improvements in the quality and quantity of data available for model building

This talk will describe a collaborative research project between Amedis Pharmaceuticals and Argenta Discovery, aiming to address both these issues:

• Amedis Pharmaceuticals has developed novel, proprietary molecular descriptors and modelling technologies, based on genetic programming and advanced statistical techniques
• Argenta Discovery is applying its in vitro ADMET expertise to generate high quality data, e.g., for inhibition of various cytochrome 450s
• Details of this collaboration and its progress towards improved models will be presented.

9:15 Predictive Modeling Techniques Being Used by Companies
Dr. Ravi Mallela, Recommind Inc.
Predictive modeling is still in its infancy. A variety of methods have been developed in an attempt to answer the challenge. This talk will provide real life case studies that focus on the currently used methods, the advantages and disadvantages, and the conclusions we can draw from them.

9:45 Integrating ADME/Tox Prediction and Informatics
Dr. Gregory Banik, General Manager, Informatics Division, Bio-Rad Laboratories, Inc.
ADME/Tox prediction is becoming an increasingly important tool in the evaluation and prioritization of compounds early in the drug discovery process. A system will be described that facilitates the deployment of multiple simultaneous prediction approaches to maximize prediction accuracy. The applicability of this to high-throughput evaluation will be discussed, as will the informatics issues relevant to in silico prediction, along with possible solutions.

10:15 Poster and Exhibit Viewing, Refreshment Break

10:45 Whole Body Autoradiography Answers Pivotal Questions for Early ADME and Toxicology
Dr. Eric Solon, Director of Autoradiography and Animal Resources, Quest Pharmaceutical Services
Whole Body Autoradiography provides tissue distribution information and can help to identify target organs and early information regarding ADME properties. Known drug interactions can be employed in the study design and answer more specific questions. Several examples of how QWBA has been used in drug discovery for ADME and Tox support will be presented.

11:15 An Ensemble Approach to Building Predictive ADME/Tox Models for Decision Support
Dr. Julie Penzotti, Director, Computational Chemistry, Rational Discovery, LLC
The development of predictive models for ADME/Tox properties provides decision-support tools that can be used in library design, lead optimization and the selection of drug candidates for development. Because multiple (possibly unknown) mechanisms can lead to the same toxicity endpoint or ADME property, algorithms for their prediction must be capable of handling multiple modes of action. Our modeling methods for ADME/Tox combine computational chemistry with an ensemble approach to machine learning, thereby selecting useful molecular descriptors and multiple predictive models to produce a single composite model that generally exhibits greater predictive power and generality.

11:45 Panel Discussion with all Morning Speakers

12:30 Luncheon

Current Challenges in Predictive Toxicology

1:40 Comments by Session Chairperson
Dr. Weida Tong, Director, Center for Toxicoinformatics, NCTR/FDA

1:45 Demonstration of Consistency in Microarray Results Across Species, Platforms and Laboratories
Dr. Andrew Cherniack, Molecular Biology Sr. Scientist Abbott Bioresearch Center, Inc.
We have used Affymetrix Genechips to construct a database containing mouse liver expression profiles induced by twelve different hepatotoxicants. We have used blinded studies to test the predictive value of the database and compared our results in mice to those obtained in rats using the Agilent microarray platform. These results are significant not only because we demonstrate the predictive value of the database tool in a blinded study but also demonstrate acceptable consistency in our microarray results across species, platforms and laboratories.

2:15 Decision Forest in the Field of Omics Land - Show Me the Confidence
Dr. Weida Tong, Director, Center for Toxicoinformatics, NCTR/FDA
Class prediction based on omics data is playing an increasing role in diagnosis, prognosis and risk assessment. It is normal that omics data has very many predictor variables (e.g. genes in microarray data or m/z peaks from SELDI-TOF mass spectrometry) over relatively few samples and, sometimes, is noisy. The nature of omics data manifests that the challenge in class prediction is no longer in constructing a fitted model that is internally consistent; rather, it is how to quantify the predictivity of the fitted model for classifying unknown samples with known certainty. We present results of a model classifying ovarian cancer using proteomic data with a novel class prediction technique, Decision Forest, where prediction confidence can be quantitatively estimated.

2:45 A MIAME for Toxicogenomics-Towards Harmonization of a New Field
Dr. Susanna-Assunta Sansone, Toxicogenomics Project Coordinator, Microarray Informatics, EMBL-EBI The European Bioinformatics Institute (UK)
Harmonization in toxicogenomics will have broad application in experimental science as well as clinical medicine. Following the very favorable response that the Minimum Information About a Microarray Experiment (MIAME) has received from the microarray community and key scientific journals, we have initiated the harmonization process as applied to array-based toxicogenomic experiments. The harmonization of toxicological data and its subsequent inclusion in public databases will significantly increase the utility of such datasets by providing the biological context of the experiment in which the microarray data was generated.

3:15 Poster and Exhibit Viewing, Refreshment Break

3:45 Predicting Drug Safety from Gene to Cell to System
Dr. Tom Colatsky, Vice President, Healthcare Research, Paradigm Genetics
Metabolic profiling (i.e. metabolomics, metabonomics) is emerging as an important new discipline focused on the comprehensive analysis of the low molecular weight biochemicals present in cells, tissues and biofluids. The ability to combine metabolic profiles with other data streams, including histopathology and pathway data, can provide additional information beyond a simple injury signal, and lays the foundation for a mechanism-based, minimally invasive approach to predicting long-term drug safety and human outcomes. This presentation will focus on use of metabolomic biomarkers to assess liver toxicity from gene to cell to system, alone and in combination with other data streams, as a means of signaling the cellular and genetics mechanisms underlying toxicity and disease.

4:15 Using In Vivo Zebrafish Assays to Predict Drug Toxicity
Dr. Chaoyong Ma, Manager, Research and Business Development,
Phylonix, Inc.
A rapid, inexpensive, reproducible and predictive vertebrate model for drug toxicity would greatly improve the efficiency of drug development. We have developed such a predictive model using the zebrafish model organism, which combines the advantages of higher throughput analysis (compared to mammalian models) and higher relevance to humans (compared to in vitro and invertebrate models). We have tested more than 60 compounds using our toxicity assays, and found that the results from our zebrafish-based toxicity tests correlated well with the toxicity data obtained from mammalian models.

4:45 Multiparametric Assessment of Drug Toxicity
Dr. Ed Luther, Principal Scientist, CompuCyte Corporation
The automation of fluorescence based analysis has brought the ability to objectively and precisely quantify the expression of molecules in predictivein-vitro test systems. This presentation will highlight the new applications made possible by advancements in laser scanning fluorescence instrumentation and the benefits being derived in predictive toxicological settings. The applicability of specific model systems will be reviewed.

5:15 Panel Discussion with all Afternoon Speakers

5:30 Conclusion of Day One

Thursday, November 20, 2003

8:30 Light Continental Breakfast

Modeling

9:00 Comments by Session Chairperson
Dr. Robert D. Clark, Tripos, Inc.

9:10 Structure-based Modeling of DM-PK and Toxicological Properties: Accurate Data, the Right Data
Dr. Robert D. Clark, Tripos, Inc.
There is currently a great deal of interest in creating computational tools for predicting drug metabolism and pharmacokinetic (DM-PK) properties such as pKa, solubility, bioavailability and toxicity. The limiting factor in most cases is the lack of suitable data from which to construct training sets. Large data sets are typically required for generalizable models because of the large number of different ways a compound may behave well or badly. Unfortunately, in cases where large amounts of data are available, they are not of the right kind. In addition, the uneven quality of data can mean that larger training sets actually give less predictive models than smaller data sets do. Such considerations make it as important to examine the available data carefully so as to avoid over-interpretation of the models obtained as it is to minimize outright errors in prediction. The complexities likely to be encountered will be discussed in general terms and illustrated in the particular by SIMCA-based molecular field analysis of hepatotoxicity data obtained in vitro.

9:40 9:40 PharmGenix™ - Combinatorial Rat Panels for Predictive Toxicology
Dr. Howard Jacob, CSO, PhysioGenix, Inc.
A major reason for failure in drug development is unforeseen toxicity detected during clinical trials. Proof of concept for the PharmGenixTM rat panel shows that the lack of genetic diversity found in commonly used animal models and cell systems can explain why current safety testing procedures often fail to predict adverse drug effects in human populations. PharmGenixTM rats are bred in a combinatorial fashion from selected inbred strains such that the resultant F1 progeny are iso-genic, display experimental reproducibility like inbreds and can capture up to 80% of the known rat genetic diversity. In practice, those com-pounds that pass safety testing in all PharmGenixTM strains are not like-ly to have adverse effects in the general human population, while detec-tion of toxicity in one strain will identify a genetic component of the adverse effect. Though in life studies, PhysioGenix demonstrates that the PharmGenixTM panel has more power to detect the nephro- and hepto- toxic actions of gentamicin and clofibrate than either the com-monly used outbred or multiple inbred strains of rat. Use of the PharmGenixTM panel therefore enables detection of the general and genetic components of toxicity that currently escape detection leading to downstream development of diagnostic and pharmacogenomic tools.

10:10 Applications of Multidimensional Filtering and Compound Redesign: Better Leads Through Balanced Property Selection
Dr. Steven L. Gallion, Research Fellow, ArQule Inc.
The process of generating quality lead series with the highest probability of success is of paramount importance to increasing the efficiency of drug discovery. Conventional properties such as novelty, potency and selectivity must now be balanced with physiochemical and related ADMET properties early in the discovery process. This can be achieved in many ways. Two examples are provided: (a) using a series of predictive properties to rank-order and score multiple compound sets for further advancement, and (b) alleviating a particular ADMET problem for a compound or drug of known efficacy.

10:40 Refreshment Break

11:10 The Discovery and Application of Predictive Toxicogenomic Markers to Screen Compounds for Liver Toxicity
Dr. Michael Czar, Research Scientist, Pharmacogenomics-Collaborative Research
One of the promises of genomic technologies is the identification of toxicogenomic markers that can provide signatures to evaluate and predict toxicity of new chemical entities. Our research group has focused on the development of a Predictive Toxicogenomic Screen™ (PTS) for hepatotoxicity based upon profiles from greater than 100 toxic and non-toxic compounds. The gene expression data were analyzed using multiple statistical modeling approaches to identify marker genes and generate predictive models of specific liver toxicity produced by the reference compounds (i.e. cholestasis, steatosis, necrosis, etc.) based on histopathology findings and clinical chemistry data. This was accomplished for both in vivo tissue samples as well as for primary rat hepatocytes so that either methodology could be used for assessing compound hepatotoxicity. How predictive toxicity models from gene expression data are being employed to prioritize test compounds at early stages in development will be presented.

11:40 Use of Artificial Intelligence Methods for the Prediction of ADME Properties
Dr. Mohammed Afshar, CEO, Ariana Pharma
Ariana Pharma is developing specific ADME-T tools, in which information gathered from databases is used in combination with knowledge extracted by the system through interaction with the expert user. These tools enable the system to flag areas of chemical space absent from the learning database and prompt the expert user for the missing data. The user can then modify the knowledge base from other sources or acquire the missing data experimentally. Another advantage of the machine learning system is that it can always explain, in a format understandable by the biologist/chemist, the reason for a particular decision and its estimated accuracy. The platform combines these tools with virtual screening methods.

12:10 Interactive Panel Discussion with all Morning Speakers

12:30 Lunch on your own

In Silico Strategies and Data Mining

1:40 Comments by Session Chairperson
Dr. Alex Aronov, Investigator, Applications Modeling, Vertex Pharmaceuticals

1:45 Screening In Silico for HERG K+ Channel Blockers
Dr. Alex Aronov, Investigator, Applications Modeling, Vertex Pharmaceuticals
Acquired long QT syndrome (LQTS) occurs frequently as a side effect of blockade of cardiac HERG K+ channels by commonly used medications. A large number of structurally diverse compounds have been shown to inhibit K+ current through HERG. There is considerable interest in developing in silico tools to filter out potential HERG blockers early in the drug discovery process. We will describe a binary model that combines a 2D topological similarity filter with a 3D pharmacophore ensemble procedure to discriminate between HERG actives and inactives with an overall accuracy of 82%, with false negative and false positive rates of 29% and 15%, respectively. This model should be generally applicable in virtual library counterscreening against HERG alone or in combination with existing high throughput HERG screening technology.

2:15 In silico - in vitro approaches to estimate in vivo ADME parameters
Dr. Jacques Migeon, Sr. Scientist, CEREP, Inc.
The BioPrint(.) dataset is a training set for computational tools that predict key in vitro and in vivo activities based on chemical structure.
Over 2200 drug-like molecules have been systematically tested across a panel of over 150 in vitro assays. These data are complemented by human pharmacokinetic and adverse drug reactions data collected from clinical literature. BioPrint tools include quantitative structure-activity (QSAR) models that estimate in vitro ADME properties and in vivo end-points such as oral absorption. Systematic analysis of the data also supports the interpretation of in vitro assay results vis-à-vis relevant in vivo endpoints. These tools can be applied at early stages of discovery to guide library design and lead selection, enriching the pipeline with compounds likely to be successful in development.

2:45 A Novel Pattern Recognition Method for Microarray Data Analysis
Dr. Huixiao Hong, Manager, Bioinformatics Lab, NCTR/FDA
DNA microarray technology enables the expression of thousands of genes to be simultaneously evaluated. Class prediction methods are widely employed to use data from multiple population classes (e.g., diseased versus normal) to develop models that can predict the class to which an unknown sample belongs, based solely on its expression profile. Most such classification methods in the literature require a priori selection of a small set of informative genes in order to make the model development computationally tractable. Using these pre-selected genes first for model training and then for cross validation is known to substantially underestimate the prediction error. We present a novel class prediction method named Decision Forest that embeds gene selection in the model development, eliminating the error rate underestimate in cross-validation.

3:15 Refreshment Break

3:45 Predicting the Toxicity and Mechanism of New Drug Candidates
Dr. Kurt Jarnagin, Vice President, Biology & Chemogenomics, Iconix Pharmaceuticals, Inc.
Data mining is critical to achieve meaningful results in predictive toxicology. However, raw data is most valuable when linked with key outcomes. By using a chemogenomic approach that couples pharmacology and toxicology with genomics tools, Iconix has developed a system that helps companies effectively 'weed out' those compounds that are likely to fail in the clinic. Using a reference collection of more than 550 drugs, failed relatives, standards and toxicants profiled using expression studies, pharmacology and traditionally toxicological end-points, we have identified biomarker sets that effectively predict mechanism of action and side effects including forms of liver, kidney, heart and marrow toxicity. By evaluating the response of many genes and their products in parallel, pharmaceutical companies can optimize their discovery processes.

4:00 Structural Alerts for Hepatotoxicity
Dr. William J. Egan, Staff Investigator, Vertex Pharmaceuticals
In the period 1960-1999, 26% of marketed drug withdrawals were caused by hepatotoxicity. We have created a structural database of hepatotoxic drugs and chemicals containing over 240 molecules, including 54 drugs which were withdrawn or whose use was abandoned due to hepatotoxicity. Besides name and structure, the database contains information on marketing status, type of injury, mechanism of injury, metabolism, and pertinent references. Analysis of the structural features, clinical and biochemical literature, and routes of metabolism shows that there are a number of structural features that are definitely or possibly associated with hepatotoxicity in humans. The presentation will summarize the database, review some of the identified structural features, and discuss the challenges involved in extracting useful alerts and rules applicable to the complex processes of liver toxicity.

4:30 Panel Discussion with all Afternoon Speakers

4:45 Conclusion of Predictive Toxicology Conference

Lead Sponsoring Publication:
Sponsoring Publications:
Web partners:

Travel Information
Special Airline Discounts Available
Special Zone and Discount Fares have been established for this conference with United Airlines. Please call United Airlines Meeting Reservation Desk at 800-521-4041 and reference ID#579YS.

Hotel Information
Boston Park Plaza Hotel
64 Arlington St.
Boston, MA 02116
T: 617-426-2000 o F: 617-426-5545
RoomRate: $159/single o $179./double
Cut-off Date: October 22, 2003

Please call the hotel directly to make your room reservation. Identify yourself as a Cambridge Healthtech Institute conference attendee to receive the reduced room rate. Reservations made after the cut-off date or after the group room block has been filled (whichever comes first) will be accepted on a space-and-rate-availability basis. Rooms are limited, so please book early.