CINF 69 |
| Pharmacophores were originally defined as distributions of generalized features in space that are required for activity against a particular biochemical target. Every active compound must exhibit every component feature in such an “essential” pharmacophore, which has the virtue of making clique detection and other deductive approaches to pharmacophore elucidation feasible. Unfortunately, the corresponding 3D search queries are often comprised of too few features too closely spaced to effectively discriminate between active and inactive ligands - i.e., queries that have a large false positive "hit" rate. But such queries can also be so complex that they are too specific. Searches based on them will only recover ligands very similar to those in the training set and will have a prohibitively high false negative hit rate. Including partial match constraints in a query makes it possible to strike a useful balance between these two extremes, especially when a mix of stringent and permissive constraints is used. In the former, most or all features are required to "hit", whereas in the latter only a few are required. This talk will describe a model for predicting the search specificity of a query from its geometry, feature composition and constituent partial match constraints. |
|
Advances in 3-D Pharmacaphores and 3-D Searching
2:00 PM-5:30 PM, Tuesday, March 27, 2007 Hyatt Regency McCormick -- 12 B, Oral
Division of Chemical Information |