SimBioSys Inc. > Science > Global Optimization of Molecular Structures

[Product Releases]

[News]

Can we trust docking results?
Sept 2010

IBM Systems and Technology Group releases a white paper with eHiTS and Cell
Oct 2008

EPA's ToxCast^TM project will use SimBioSys' eHiTS as docking engine
Nov, 2007

[Events]

240th ACS
Aug 22-26, 2010
Boston, MA, USA
booth #945
see >> more

Global Optimization of Molecular Structures

The problem:

In many structural biology problems, we are dealing with a global optimization problem. In such problems the search space consists of molecular conformations and the goal function includes the conformational and binding energies. Examples of such problems are:

Protein-Ligand Docking. In which the conformation of the ligand and its position relative to the protein binding site should be optimized. If the protein is not rigid the conformational space of the protein should also be included.
Protein-Protein Docking. This is similar to the protein-ligand docking problem but the conformational space is usually more constrained because of the size of the two molecules.
Protein Folding. In which the conformational space of a protein is searched, looking for the minimum energy conformation.
Side-Chain Packing. Which is a sub-module of some of the protein folding methods. The protein backbone is fixed in this problem and the goal is to place the side chains with minimum energy.
Organic Crystal Structure Prediction. In which the goal is to predict the packing of drug-size molecules at lowest crystal lattice energy level.

One of the main issues in solving all these problems is that the energy functions of such molecular structures is a high dimensional non-linear function with many local minima. There are many stochastic approaches for solving such problems but in SimBioSys we believe in exhaustive search which can guarantee certain accuracy.

The SimBioSys solution:

The general idea behind our search strategies can be explained as follows:

Perform an exhaustive sampling of the search space.
Select a subset of sampled solutions based of geometric diversity and goal function value.
Locally optimize the selected points.

These steps are illustrated for a hypothetical energy landscape shown in the picture. The white grid represents the sampling phase and the black arrows show the local optimization step.

We have successfully applied this search strategy to some of the aforementioned problems, namely protein-ligand docking and crystal packing, and developed a wide range of tools that can be adopted to solve similar problems.

The main bottleneck in applying this approach is its processing time requirement. With many years of experience in improving the efficiency of such methods, we have adopted state of the art methods in computational geometry, algorithm design, and numerical optimization to achieve good performance there.

Home | Site Map | Legal Disclaimer