19th-banner-rev.gif

Book of Abstracts: Albany 2005

category image Volume 22
No. 6
June 2005

Predicting Molecular Association

Computational prediction of the binding geometry and affinity for various protein ligands is an area of critical importance considering the rapidly growing number of known protein structures. We focused on several related computational tasks:
  1. predicting small molecule binding sites from a single structure;
  2. predicting protein-protein interaction patches from a single structure
  3. predicting conformations of protein loops with high accuracy
  4. docking flexible ligands to flexible protein pockets
  5. predicting association of two proteins from their uncomplexed structures
We demonstrated that the small molecule binding pockets can be predicted with a certain transformation of the Lennard Jones potential. This prediction does not depend on the ligand and captures the majority of the actual binding sites (96% of all known pockets at the 50% overlap level) (1). The predicted envelopes were classified and hierarchically clustered to form what can be called a pocketome. This algorithm can be useful to predict new or allosteric binding sites or the feasibility of inhibiting protein-protein interaction with a small molecule.

Predicting transient protein-protein interaction interfaces without known the partner is not an easy task because the interaction surfaces look very similar to the non-interfaces. Both physical methods and methods based on sequence conservation were proposed. We improved our previously published method (2) based solely on the physical surface descriptors, optimized it and tested it on a large benchmark of protein interactions. Surprisingly, this signal is sufficiently strong and adding the sequence conservation map does not improve the prediction significantly. Therefore, the protein interaction propensity can be predicted even if no sequence alignment is available.

Predicting conformations of protein loops is an important task in several areas, e.g. homology modeling and simulating the association-induced conformational changes. We introduced an improved conformational sampling protocol that predicted loops of up to 15 residue long with high crystallographic resolution accuracy (under 0.5 Å for the backbone atoms and under 1 Å for the side chains). The protocol was tested on a large benchmark of loops.

Including the receptor pocket flexibility into account in ligand docking remains to be a task highly specific to the nature of the receptor. We attempted formulate a docking protocol which is relatively general and includes both the side-chain sampling and loop movements (3). This protocol can be further enhanced and generalized if the backbone is sampled according to the relevant normal modes.

The CAPRI protein docking experiment provides a good platform for evaluation of protein docking algorithms. While a fully reliable prediction of the association geometry given two uncomplexed structures remains unattainable, considerable progress has been made the last few years (4). The difficulty of the problem is a function of the scale of the induced conformational changes upon association. We demonstrate that at least in some cases these changes can be correctly predicted.

References and Footnotes
  1. An, Totrov, and Abagyan. Proteomics (in press), 2005.
  2. Fernandez-Recio, Totrov, Skorodumov, and Abagyan. Proteins, 58,134-143 (2005).
  3. Cavasotto and Abagyan. J Mol Biol 337, 209-225, (2004).
  4. Fernandez-Recio, Totrov, and Abagyan. Proteins 52 113-117, (2003).

Ruben Abagyan*
Jianghong An
Claudio Cavasotto1
Juan Fernandez-Recio
Irina Kufareva
Maxim Totrov1

The Scripps Research Institute
10550 N Torrey Pines
La Jolla, 92037, USA
1Molsoft, LLC

*Phone: 858-784-8595
Email: abagyan@scripps.edu