News and Publications

Computational Structural Genomics and Molecular Design

R. Abagyan, J. An, A. Bordner, L. Brive, C. Cavasotto, H.S. Choi, J. Fernández-Recio, V. Katritch, W.H. Lee, B.D. Marsden, A. Orry, M. Totrov,* M. Schapira,* Y. Zhou,** M. Dawson,*** X. Zhang***

* MolSoft LLC, La Jolla, CA
** Genomics Institute of the Novartis Research Foundation, San Diego, CA
*** The Burnham Institute, La Jolla, CA

HOMOLOGY MODELING

Building models of proteins via homology bridges the growing gap between sequence and structure. Threading or sequence-structure alignment is the first key procedure in the process. We derived a "standard of truth" database of structural alignments based on known high-resolution x-ray structures. This database has been used to derive and optimize the parameters of a sequence-alignment method that takes into account the predicted and actual secondary structure of the sequence and the templates, respectively, and the degree to which each residue is buried in the template structures. We successfully applied this new method in a comprehensive assessment of the homology modeling approach that uses the Internal Coordinate Mechanics (ICM) software with other methods such as Modeller and Swiss-Model.

One use of homology models is as search models for molecular replacement calculations. Preliminary results indicate that predicting local reliability of a protein model yields better results for homology models with less than approximately 35% sequence identity, increasing the number of instances in which molecular replacement is successful. We were able to increase the overall success rate of the molecular replacement procedure and generate interpretable electron density maps in previously uninterpretable cases.

PROTEIN-PROTEIN INTERACTIONS AND DOCKING

Predicting protein-protein interactions by docking the protein subunits is a major challenge in computational biology because of a marked induced fit of the interface when a complex is formed. To address these problems, we continued optimizing and extending our protein-protein docking method based on the ICM global energy optimization. The use of "soft" potentials precalculated in a 3-dimensional grid, previously parameterized for proteins, markedly improved the efficiency of the procedure (Fig. 1).

The method was recently validated in a benchmark set of 24 protein-protein complexes, and its predictive efficiency was close to 100% when the complexes were rebuilt by starting from their complexed subunits. When the unbound structures of the subunits (i.e., in more challenging and realistic conditions) were used, use of the method resulted in correct predictions of the near-native conformation within the top 20 of a list of candidate docking solutions in 85% of the complexes with no major backbone motion upon binding. Among them, 7 (64%) of 11 protein-inhibitor complexes were successfully predicted as the highest-rank conformations, the best result achieved so far in protein-protein docking.

The predictive efficiency of the protein-protein docking procedure yielded excellent results during the recent competition of methods for prediction of protein interactions (CAPRI: http://capri.ebi.ac.uk). In particular, for 1 of the 3 targets, for which 13 of the best experts in the field proposed structures, the closest model to the crystallographic structure was proposed by our group; 71% of the intermolecular contacts were correct.

VIRTUAL LIGAND SCREENING, DOCKING, AND DRUG DISCOVERY

Flexible docking combined with virtual ligand screening is an emerging technology in rational drug design. Two novel antagonists of the retinoic acid receptor were identified by using virtual ligand screening and were experimentally tested by our collaborators.

In collaboration with X. Zhang and coworkers at the Burnham Institute, La Jolla, California, we are using homology models and the orphan nuclear receptor TR3 to determine how well we can predict small-molecule binders. Using computational approaches, we are screening hundreds of thousands of molecules into homology models of the receptor in the search for new ligands that can modulate the protein's activity as an anticancer agent. This search for new ligands is fundamentally important because only a fraction of all known protein sequences with potentially important therapeutic uses have a known 3-dimensional structure.

We are also using the ICM virtual ligand screening and docking technology to examine the structure of neurologically important proteins. Recently, we began analyzing ligand-binding pockets in G protein-coupled receptors, which are associated with neurologic diseases, and rationally docking agonists and antagonists into them. As a way of validating these experiments, we accurately identified the retinal ligand-binding pocket in the crystal structure of the G protein-coupled receptor bovine rhodopsin and can precisely dock the ligand into that region (Fig. 2).

Virtual ligand screening can be used not only to detect drug candidates but also to determine which of thousands of biological substrates is a native ligand. We used virtual ligand screening and the KEGG ligand database from Kyoto University, Kyoto, Japan, which consists of more than 6000 biological substrates, to identify the native ligand for the crystal structure of bovine rhodopsin. The screening narrowed the search to less than 0.5% of the database.

ELECTROSTATICS

Calculating electrostatic free energy of solvation accurately and quickly is critical in simulations and in the evaluation of binding affinity. Using the boundary element algorithm to solve the Poisson equation provides the most accurate calculation for an implicit model. The improved algorithm can be used in folding simulations and to calculate the electrostatic component of binding in a matter of seconds.

HIGH-DENSITY DNA CHIPS

We proposed and tested a novel algorithm to determine the level of expression on a high-density DNA chip by using a smaller number of representative oligonucleotides. This algorithm does not depend on the so-called mismatch probes and may be a strategy to at least quadruple the number of genes represented by a single chip.

PUBLICATIONS

Abagyan, R., Totrov, M. High-throughput docking for lead generation. Curr. Opin. Chem. Biol. 5:375, 2001.

Brive, L., Abagyan, R. Computational structural proteomics. Ernst Schering Res. Found. Workshop 38:149, 2002.

Fernández-Recio, J., Totrov, M., Abagyan, R. Computational prediction of protein-protein interactions and rational drug design. Curr. Top. Med. Chem., in press.

Fernández-Recio, J., Totrov, M., Abagyan, R. Screened charge electrostatic model in protein-protein docking simulations. Pac. Symp. Biocomput. 7:552, 2002.

Fernández-Recio, J., Totrov, M., Abagyan, R. Soft protein-protein docking in internal coordinates. Protein Sci. 11:280, 2002.

Schapira, M., Raaka, B.M., Samuels, H.H., Abagyan, R. In silico discovery of novel retinoic acid receptor agonist structures. BMC Struct. Biol. 1:1, 2001.

Totrov, M., Abagyan, R. Rapid boundary element solvation electrostatics calculations in folding simulations: successful folding of a 23-residue peptide. Biopolymers 60:124, 2001.

Zhou, Y., Abagyan, R. Match-Only Integral Distribution (MOID) algorithm for high-density oligonucleotide array analysis. BMC Bioinformatics 3:3, 2002.