Support the development, validation, and use of computational toxicology software of mutual benefit to the FDA/CDER and NIH/NIDA

National Institute On Drug Abuse

Investigators

Abstract

Computational toxicology, incorporating advances in computer technology, toxicology databases and (quantitative) structure-activity relationship ((Q)SAR) software, offers a rapid, cost-effective means for screening large numbers of compounds to help eliminate those with potentially unfavorable toxicity profiles to prioritize promising compounds for further testing. In drug discovery, the application of combinatorial chemistry and high throughput screening of compounds has resulted in an increase in the number of compounds identified with potentially desirable pharmacological properties. The selection of lead compounds for further development can be hampered by limitations in traditional methods for assessing toxicity. There is a continuing need for more rapid and cost-effective methods for screening large numbers of compounds for potential toxicity. Information from toxicology databases is being used by FDA/CDER to improve the performance of computational toxicology models to meet FDA/CDER needs, as well as reduce the use of test animals and to provide a screening method for lead compound selection in drug discovery. With the support of past Inter-Agency Agreements between NIH/NIDA and FDA/CDER, computational toxicology models have been developed by FDA/CDER for use with several commercial (Q)SAR software programs to predict the potential toxicological and adverse human effects of many classes of organic chemicals, including those used as drugs. These software programs, together with FDA/CDER models, can be used to rapidly estimate the toxic potential of substances for which no or inadequate experimental data are available, or for which additional evaluation might be useful. This information can be used as one source of support for regulatory decisions. Such software and models can also be of value to the pharmaceutical industry as an aid for compound selection in drug development. Computational toxicology models developed by FDA/CDER assess potential toxicological properties of test compounds by evaluating the degree of local and/or global similarity to substances with known chemical, physical and biological properties. (Q)SAR models developed and/or used by FDA/CDER encode correlations between molecular sub-structural fragments and activity at toxicological endpoints of regulatory significance. These (Q)SAR models are used to make a prediction of the toxicological potential of a new compound based solely on chemical structure by assessing the presence or absence of these same sub-structural fragments, combined with other calculated physical chemical parameters. The models used by FDA/CDER are based on large training data sets linked to chemical structures with a broad range of sub-structural attributes. Models for non-clinical endpoints are based on data from FDA/CDER approval packages, testing programs such as the National Toxicology Program, and other reliable external sources; models for clinical endpoints are based on human adverse event data extracted from post-market surveillance databases and drug labels. FDA/CDER models have high coverage for pharmaceuticals. Under Research Collaboration Agreements (RCAs) with three computational toxicology software developers, FDA/CDER continues to expand training databases used for (Q)SAR modeling and develop new models with improved predictive performance and structural coverage. The FDA/CDER toxicity and human adverse effect models are now commercially distributed by two RCA partners and are being use by pharmaceutical companies for lead selection in drug discovery.