CINF 80 |
| We report a virtual screening protocol for the identification of identifying diverse HIV inhibitors. We developed linear and non-linear classification models based on a 900 compound training set. The models were then used to predict the activity class of a large vendor library. The vendor compounds that were predicted active were then filtered based on similarity to the most active compound in the training set. The final hit list was prioritized according to similarity to the most outlying active compound in the training set. Our initial results did not lead to a significantly diverse set of hits. Furthermore, none of our hits were common to the set obtained previously using a pharmacophore model. However, relaxing the similarity constraints identified four compounds which were very similar to a known inhibitor. We discuss possible reasons why this was the case and describe docking results for the hits obtained using our tiered screening protocol. |
|
Advanced Mining and Use of Life Science Information
8:25 AM-12:00 PM, Wednesday, March 28, 2007 McCormick Place North -- Room N134, Level 1, Oral
Division of Chemical Information |