MOUNTAIN VIEW, Calif., May 4 /PRNewswire/ -- Today, Iconix Pharmaceuticals announced the publication of a paper that describes the analysis and generation of gene expression based signatures to uncover the underlying mechanisms of drug toxicity. The paper, titled "Classification of a large micro-array data set. Algorithm comparison and analysis of drug signatures" compares a variety of proprietary and non-proprietary mathematical algorithms that focus on creating gene signatures that can predict the toxicological properties of drugs. After examining multiple linear and non-linear algorithms those that were selected combined two key attributes: predictive power and a gene list short enough to be used in a robust diagnostic device. Significantly, the data used for this comparison doubles the amount of publicly available toxicogenomic information in the rat, the reference species used by toxicologists.
The algorithms were assessed using DrugMatrix, the world's largest database of chemogenomic data, composed of over 13,000 microarrays and encompassing the response of rats to 630 failed and approved drugs and environmental toxins and standards. One use of such a database and the signature technology described in the paper is to identify toxic liabilities of compounds early to help prioritize candidate compounds in the pre-clinical stage of development.
"Biologists want to know, as early as possible, whether or not their sample shows signs of toxicity," said Georges Natsoulis, Sr. Director of Advanced Technology at Iconix Pharmaceuticals and lead author on the paper. "But they also want to know why. The algorithms we describe in the paper combine high performance classification with interpretability, and the user requires both attributes in order to extract biological information from the analysis of gene signatures." Natsoulis went on to note, "These signatures are composed of very few genes which is an added bonus for diagnostic device development."
The analysis focused on a family of new linear classification methods including Sparse Linear Programming (SPLP) and Sparse Logistic Regression (SPLR) and their performance classifying drugs in four categories: fibrates, statins, azoles and a group of heterogeneous toxicants.
The paper is published in the May issue of Genome Research.
The manuscript is available at:
The lead author was Georges Natsoulis of Iconix Pharmaceuticals.
The Department of Electrical Engineering and Computer Science at UC
Berkeley and SPSS Inc, Chicago, as well as additional Iconix scientists,
are also authors.
About Iconix Pharmaceuticals:
Iconix Pharmaceuticals, Inc. is pioneering the new field of toxicogenomics, the integration of chemistry and genomics to profile drug candidates. Iconix's toxicogenomic capabilities enable pharmaceutical companies to increase the odds of advancing the right compounds to the clinic, reducing attrition rates and the costs of drug discovery. Iconix provides reference systems and know-how to predict toxic liabilities, side effects and mechanisms of drug candidates. The company has strategic partnerships with an ecosystem of leading life sciences companies, including MDS Pharma Services, Incyte Corporation and GE Healthcare, and collaborations with Bristol Myers Squibb, Abbott Laboratories, Eli Lilly, Schering-Plough, AstraZeneca, Eisai Co., Ltd. and Millennium Pharmaceuticals, Inc. Iconix also provides research, training and support to the U.S. Food and Drug Administration, Center for Drug Evaluation and Research (CDER) under an agreement to advance CDER's study of the application of genomic technologies in the regulatory approval process. Iconix's DrugMatrix system has been installed at the FDA for use by CDER scientists and reviewers in toxicogenomics applications. Headquartered in Mountain View, California, Iconix was founded in 1998 and is privately held. For more information, visit http://www.iconixpharm.com/.
CONTACT: Alan Engelberg of Iconix Pharmaceuticals, +1-650-567 5527, or firstname.lastname@example.org; or Jennifer Larson for Iconix Pharmaceuticals, +1-415-409-2729, or email@example.com.
Iconix Pharmaceuticals, Inc.