next up previous contents
Next: Refraction tomography Up: Results and related Previous: Image guided surgery

Molecule energy minimization

The conformation of a molecule is a description of its 3-dimensional structure. The physical properties and biological functions of organic molecules are intimately related to their conformations. Thus computing energetically stable conformations of organic molecules is an important problem. One way to compute the conformations is to minimize an energy function, but unfortunately, the energy function typically has many local minima.

  
Figure: The Cycloheptadecane molecule has 51 atoms. The three coordinates for each atom are the input to the objective function. Thus the problem has 153 dimensions.

Procedure 2-1 was tried on a small instance of this problem. Cycloheptadecane is a macrocyclic molecule. It consists of a chain of 17 carbons, where each carbon is connected to two hydrogen atoms as well as two other neighbor carbons. Figure gif illustrates this structure. There are a total of 51 atoms in the molecule. Treating their x, y and z coordinates as separate parameters of the energy function, we have an optimization problem of 153 dimensions.

Recently, a novel method to solve this problem efficiently was proposed [Wang, 1994]. First, approximate solutions are found by combining smaller chains together. Then, these approximate solutions are further refined in an optimization loop. Figure gif contains the comparison of Procedure 2-1 with Powell's method in this loop as well as a sample run starting from a random initial conformation.

Note that the gradient of the energy function typically can be computed. In that case an algorithm that can use this information such as conjugate gradient [Press et al., 1992,Polak, 1971], will be more efficient. The comparison is made with Powell's method because it is known as one of the most efficient algorithms among the ones that do not use the gradient.

  
Figure: Comparison of Procedure 2-1 with Powell's method on the molecule energy minimization problem. The first example starts from a near optimal configuration. The second example starts with a random configuration. The lower lines are the traces of Procedure 2-1.



next up previous contents
Next: Refraction tomography Up: Results and related Previous: Image guided surgery



Deniz Yuret
Tue Apr 1 21:38:29 EST 1997