|
|
|
|
|
|
|
|
|
|
| John Irwin is a research scientist in the Department of Pharmaceutical Chemistry at the University of California, San Francisco in the group of Brian Shoichet. Dr. Irwin received his PhD in Organic Chemistry at the ETH in Zurich in 1991. After working briefly at a startup company developing molecular modeling software, he joined the lab of Gerard Bricogne at the MRC Laboratory of Molecular Biology in Cambridge, England developing new methods of macromolecular structure determination. He developed one of the early web-based graphical interfaces in structural biology. He went on to join the Macromolecular Structural Database project at the EMBL-EBI also in Cambridge where he was part of a team transforming the Protein Databank into an Oracle database and developing research tools. He joined the Shoichet Lab while still at Northwestern in 2000 and moved to UCSF in 2003. His work focuses on improving virtual screening methods, and on lowering the barriers to entry to docking for non-specialists. Dr. Irwin has consulted on drug discovery projects for a number of pharmaceutical companies.
|
|
Investigating bias in docking screens with target, ligand and decoy benchmarking sets
John Irwin, UCSF
N. Huang, B. K. Shoichet and J. J Irwin*, University of California, San Francisco
* presenting author
Structure-based virtual screening is now the most widely-used approach to leverage structure for ligand discovery. Whereas the ultimate test of docking is prospective prediction of novel ligands, a pragmatic approach for routine testing of algorithmic developments is to use experimentally-observed poses of selected ligands and enrichment of known actives as performance evaluation criteria. Here we report datasets that may be used to benchmark docking programs. We are making available a set of actives drawn from the literature and a corresponding Database of Universal Decoys (DUD) for forty drug targets. Decoys were selected from commercially-available, ‘drug like’ compounds to have similar physicochemical properties to known actives, while having dissimilar chemical structure. To facilitate routine testing of our docking program against these forty systems we developed a high throughput virtual screening pipeline. We have docked DUD and its actives forty drug targets to investigate the performance of our docking program, our automated docking procedure, and the database itself. We have also compared docking against DUD with other databases such as the MDDR. Our results show that enrichment depends on the database used for docking, and suggest strongly that a carefully calibrated decoy database is important for effectively evaluating docking enrichment. To control for interference between annotated lists for different targets we report cross-docking experiments for each of the 40 systems. We are making DUD and the database of actives available for free download in ready-to-dock formats in the hope that DUD will be a useful community resource for improving docking methods.
|
|
|
|
|
|
|
|
|
