Sampling molecules
 

 
While finding all the chemical structures that match a given set of physico-chemical analytical data seems to be an intractable computational problem, it is possible to use stochastic techniques to generate a sample of three-dimensional molecular models that is statistically representative of the entire population of potential models.

Performing a structure elucidation using a stochastic approach is similar to studying the conformational space of a molecule using stochastic methods such as Monte Carlo or Genetic Algorithm methods.  However, in the case of structure elucidation the search space is no longer composed of the ensemble of all possible conformations, but is composed of the finite number of possible structural isomers that can be constructed from a set of analytical data.

We have demonstrated that by using a stochastic approach, it is possible to generate a sample of three-dimensional molecular models that statistically represents the entire population of all the possible models that can be built from a set of analytical data.
 

We have also designed a stochastic algorithm based on the simulated annealing method that searches chemical structures with desired properties.  Theoretically, the algorithm was shown to be efficient (i.e., polynomial-time).  Practically, the algorithm performs remarkably well even in very large search space sizes (up to 1032).

The above sampling techniques are being applied to investigate molecular structures studied in polymer science, biochemistry, geochemistry, fuel,  petroleum, and materials sciences.  The proposed stochastic techniques appear to be essential tools for anyone who wants to use molecular modeling techniques for structurally unresolved molecular compounds.
 


Papers

 

  Papers (applications)
 

 
Page maintained by Jean-Loup M. Faulon , last updated 10/14/97