November 10, 2016
The approach used by regulators to initially screen new chemical products for toxic effects is wrong almost half the time, according to scientists at the University of North Carolina at Chapel Hill. They have proposed an improvement that could increase accuracy to as much as 85 percent, saving millions of dollars and years of development time for new drugs and other products while improving safety.
Regulatory agencies, such as the Food and Drug Administration and the Environmental Protection Agency, that are charged with evaluating new drugs and other chemical products rely on an initial screening of the product’s molecular structure. Any groups of atoms that are believed to be linked to chemical toxicity trigger a structural alert. A product that generates a structural alerts is sent back for more testing.
Researchers led by Alex Tropsha, Ph.D., the K. H. Lee Distinguished Professor at the UNC Eshelman School of Pharmacy, determined that structural alerts are accurate in predicting toxicity only about 50 to 60 percent of the time. They developed a computational approach that uses statistical analysis to determine how trustworthy an alert is. Their improvement augments the simple-but-often-wrong thumbs up or thumbs down currently provided.
“A lot of chemicals are wrongly identified as potentially toxic even though in the end they are not toxic and that could have been predicted,” Tropsha said. “Companies are forced to run a lot of unnecessary and costly experiments, and because companies run these checks themselves before submitting their products to regulators, there are products that never see the light of day because they are flagged as toxic when they are not.”
Many existing medications trigger toxicity alerts under current testing systems that would prevent, or at least delay, bringing them to market, Tropsha said. For example, Lipitor, the best-selling drug of all time, has five elements in its molecular structure that are flagged as structural alerts but is not toxic.
By layering a technique called quantitative structure-activity relationship, or QSAR, modeling over the existing alerts system, the UNC researchers can account for the structure of the entire chemical molecule and assign a numerical value to the chance that an alert is accurate. Their innovative strategy is published in the journal Green Chemistry.
“Structural alerts are a convenient system, but there are few consequences for being wrong even though the stakes are potentially very high,” Tropsha said. “If the alert is right, then it’s ‘we told you so.’ If it’s wrong, ‘well, it was just a warning anyway.’ But unfounded alerts unnecessarily add years and millions of dollars to the cost of bringing a new drug or product to market without improving safety. That is unacceptable, we think.”
Tropsha’s group plans to make their system freely available to regulators and scientists as web-based computer software.
“We want to alarm regulators that structural alerts overpredict toxicity while missing truly toxic substances and offer them much more accurate tools to support regulatory decisions,” Tropsha said.
Study Authors and Funding
This study was supported in part by a grant from the National Institutes of Health (GM096967).
- Vinicius Alves, Laboratory for Molecular Modeling, UNC Eshelman School of Pharmacy, and Laboratory for Molecular Modeling and Design, Department of Pharmacy, Federal University of Goias, Brazil
- Eugene Muratov, Laboratory for Molecular Modeling, UNC Eshelman School of Pharmacy, and Department of Chemical Technology, Odessa National Polytechnic University, Ukraine
- Stephen Capuzzi, Laboratory for Molecular Modeling, UNC Eshelman School of Pharmacy
- Regina Politi, Laboratory for Molecular Modeling, UNC Eshelman School of Pharmacy
- Yen Low, Netflix, San Francisco
- Rodolpho Braga, Laboratory for Molecular Modeling and Design, Federal University of Goias, Brazil
- Alexey V. Zakharov, National Center for Advancing Translational Sciences, National Institutes of Health
- Alexander Sedykh, Sciome LLC, Research Triangle Park, N.C.
- Elena Mokshyna, Laboratory of Theoretical Chemistry, A.V. Bogatsky Physical-Chemical Institute NAS of Ukraine
- Sherif Farag, Laboratory for Molecular Modeling, UNC Eshelman School of Pharmacy
- Carolina Andrade, Laboratory for Molecular Modeling and Design, Department of Pharmacy, Federal University of Goias, Brazil
- Victor Kuz’min, Laboratory of Theoretical Chemistry, A.V. Bogatsky Physical-Chemical Institute NAS of Ukraine
- Denis Fourches, Department of Chemistry and Bioinformatics Research Center, North Carolina State University
- Alexander Tropsha, Laboratory for Molecular Modeling, UNC Eshelman School of Pharmacy