COMP 65 |
| New genomic technologies, using small molecules as probes for phenotypic screening, promise to reveal larger system relationships, but identifying the individual macromolecular effectors is still a difficult and time consuming task. To relate a specific compound with its biological space we have used a database of >46,000 diverse small molecules having measured activity toward 546 target proteins to test 3 different methods of relating chemical structure to biological function. Both 2D and 3D methods of similarity searching were evaluated for predicting a molecule's biological target with success rates up to 90%. If close analogues are present in the database, small molecule targets were more reliably predicted by 2D descriptors. However, the incorporation of 3D descriptors is shown to enhance the procedure by accurately assigning target activity for molecular probes with structural dissimilarity to other entries in the reference database. Additionally, the “fuzzy” FEPOPS 3D approach tested here provides automatic alignments of the biologically important bits present in the chemical structures. We use crystal structures of protein-ligand complexes to show that the predicted alignment of chemical features correlates with the biological mode of binding. |
|
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 |