DRUG DEVELOPMENT Classifying Drugs by Gene Expression Analysis

In diseases with unknown molecular etiologies, defining a high-throughput screen for drug development is particularly challenging. Gunther et al. describe a screen based on gene expression profiling by microarray analysis to classify and identify drugs used to treat depression or psychosis and those that interact with opioid receptors, which are relevant for pain management and addiction. They used two supervised classification methods (both of which classify unknowns based on existing classification of known samples): the classification tree (CT) and random forest (RF) schemes. Primary cultures of human neuronal cells were treated with antidepressants belonging to each of the four known classes, atypical, tricyclic, monoamine oxidase inhibitors, and serotonin selective reuptake inhibitors (SSRIs); classical or atypical antipsychotic agents; or opioid receptor agonists for the -, , or µ opioid receptors. Gene expression was analyzed by microarray technology and only those genes that were stimulated at least 3 times over background were included in the subsequent analysis. CT was effective in classifying the drugs into their clinically relevant uses 89% of the time and RF was effective 83% of the time. Furthermore, RF analysis also correctly categorized SSRIs and tricyclic antidepressants into their subclasses when these were handled as unknowns in the model. Their results and methods raise several interesting issues: (i) the genes identified as representative for each drug class may provide novel targets and insights into the molecular mechanism underlying the disease; (ii) the misclassified drugs may be candidates for therapies previously unknown, for example the -opioid receptor agonists were misclassified as antidepressants by both CT and RF, which suggests that these receptors may be relevant targets for antidepression treatment; and (iii) novel classes of drugs may be discovered, in particular, new drugs for complex disease phenotypes where multiple convergent pathways may contribute to the disease state.

E. C. Gunther, D. J. Stone, R. W. Gerwien, P. Bento, M. P. Heyes, Prediction of clinical drug efficacy by classification of drug-induced genomic expression profiles in vitro. Proc. Natl. Acad. Sci. U.S.A. 100, 9608-9613 (2003). [Abstract] [Full Text]

The editors suggest the following Related Resources on Science sites:

VIRTUAL JOURNAL
Prediction of clinical drug efficacy by classification of drug-induced genomic expression profiles in vitro

Erik C. Gunther, David J. Stone, Robert W. Gerwien, Patricia Bento, and Melvyn P. Heyes (5 August 2003)
PNAS 100 (16), 9608. [DOI: 10.1073/pnas.1632587100]
 |  Abstract »  |  Full Text »  |  PDF »