Ruben Abagyan - The Scripps Research Institute: "Structures, Drugs and Computing". Talk at the UCSD Bioinformatics Symposium

The focus of this symposium was "most important open problems in bioinformatics." Dr. Abagyan represented the structure community and discussed docking and modeling.

He apparently has an affinity for -omes. There is a need to characterize the "pocketome", meaning a method for finding meaningful pockets in the PDB for docking small molecules. Using an approach to do this they create the "deorphanome" which can find the native substrate of a ligand in silico. This uses known receptor structures in the PDB and matches them to a particular substrate from KEGG. They find the approach works pretty well with a couple limiting factors: the prediction accuracy depends largely on the resolution of the atomic structure they are docking to and the speed of docking. Currently, it takes about 20 seconds to dock per compound for structure; they need intelligent ways to filter out ligand substrate pairs that have no chance of working.
They have also made an attempt to keep their work from becoming simply theoretical. When looking for receptors for the androgen molecule they found a dock to a known anti-psychotic protein: they found that it binds to the androgen receptor in silico (this was not known) and indeed it was an inhibitor experimentally. Then changed the molecule a bit to not have this antipsychotic effect. This was a nice example of getting close to rational drug design.
He also described work in predicting loop regions (the most disordered regions) of protein structure. They find that up to 12 residues they can model under .5 A which is within crystal accuracy. In predicting loop regions using a monte carlo approach they find that at 120 hours they get the right dock. It's just a matter of devoting enough computational time to the problem...