CINF 13 |
| Compound similarity-based virtual screening experiments have been conducted using a variety of different drug targets, 2D and 3D descriptors, and ranking approaches. Particular attention has been paid to assembling data sets such that each active compound represents its own unique chemotype. This condition guarantees that a similarity recognition event between active compounds constitutes a scaffold hopping event at the same time. In a series of virtual screening studies involving 7 drug targets with the number of actives varying between 4 and 13 and 9969 MDDR compounds as negative controls it has been found that atom pair descriptors, SciTegic fingerprints, and 3D pharmacophore fingerprints combined with ranking, voting, and consensus scoring strategies perform well in finding new bioactive scaffolds. The performance of descriptors largely depends on the structure of the database of compounds subjected to a virtual screen. If topological biases exist between actives, as is often the case when literature data sets are used in recall experiments, 2D topological fingerprints often perform best. However, if such biases do not exist as often the case when independent compound collections are screened, pharmacophore descriptors perform well. A comparison of virtual screening performances achieved with structure-based and compound-similarity based methods will be presented also.
|
|
Computer Assisted Drug Design: Reminiscing about the Future, A Symposium Honoring Yvonne C. Martin
2:00 PM-5:20 PM, Sunday, March 25, 2007 Hyatt Regency McCormick -- 12 B, Oral
Division of Chemical Information |