|
 |
| Prediction flow chart. |
One of the biggest problems facing drug development is learning about unexpected toxicity problems late in the clinical trial process. Thus, a simple tool that would allow researchers to better predict compound toxicity is highly desirable. Recently, Jiansuo Wang and Luhua Lai at Peking University (Beijing) examined the use of data mining to assist in determining the toxicity of potential lead compounds before their synthesis (http://preprint.chemweb.com). They devised an in silico design technique thatcould enable researchers to find molecules with a higher chance of being safe in humans than if they simply picked the most active compounds.
The researchers took a two-pronged approach to mining the Registry of Toxic Effects of Chemical Substances database of 125,586 toxic chemicals (www.cdc.gov/). First, basic features such as molecular weight, atomic composition, and two-dimensional structure (based on parts of the molecules) are determined for the toxic substances in the database. Then similar structures are placed into groups by clustering and similarity computation. These patterns reveal some structure patterns of toxic compounds. The structures are then tested for quantitative structure–activity relationships (QSARs). The techniques create a balance between speed and accuracy in mining such a database.
The benefit of a database based on structure is that once QSAR is conducted on a molecule that is being considered for therapeutic use, similar structures could be identified in the database and intensive QSAR studies conducted. This would reduce the number of compounds to which the potential drug need be compared. The researchers indicate, however, that the lack of standardization of toxicology databases, which reduces the number of compounds that can be searched, hinders the effective use of predictive toxicology.
|
