COMP 64 |
| Methods for evaluating the similarity of small molecules for the purposes of drug discovery are numerous and varied. A recurring debate within the field has to do with the relative utility of 2D versus 3D methods for virtual screening or for ligand design exercises. One particular challenge facing this field is a serious complication of inductive bias. Data sets are highly variable with respect to the degree to which test molecules are trivially similar to query molecules. Segregation of training and test ligands can be made randomly, but given that chemical exploration often elaborates chemical classes with minor substituent variation, this leads to tests where most methods perform equivalently. Segregation based on temporal or intellectual property relationships provides a means to test similarity methods effectively in the way in which they must be used in practice. A number of data sets and the performance of multiple methods will be presented. |
|
Molecular Similarity and Indexing Methods
9:00 AM-12:00 PM, Monday, 11 September 2006 Moscone Center -- Room 224/226, Oral
Division of Computers in Chemistry |