Minimizing the Energy of the Alanine Dipeptide by Simulated Annealing

This paper presents an approach for deriving 3-D structures of polypeptide chains which have minimum energy. The well-known optimisation algorithms are usually applied on the model which contains hard constraints over coordinates of the atoms. The drawback of such strategies is an ineecient search for the optimal solution. Our approach proceeds in two steps: rst, the standard model is transformed into an equivalent model without hard constraints, and second, the simulated annealing local search algorithm is performed. The empirical results demonstrate that simulated annealing on the proposed model outperforms traditional search algorithms especially with respect to running times. In the pharmaceutical industry new methods for predicting protein structures and engineering proteins are used in the development of new`designer drugs'. Despite the fact that the 3-D structures of 500 proteins are known, one is still unable to formulate general rules to predict the structure of proteins which are built from the amino acid sequences of polypeptide chains. There are more than 50000 known proteins and the number of such proteins grows much faster than the number of proteins with determined 3-D structure. The most widely used method to compute protein structures is X-ray diiraction 7], the main drawbacks of this method are its cost, the time needed to compute a structure (several years) and its lack of generality to all proteins because of the crystallisation problem 4]. Using computer-based approaches specialised software tools have been considered to assist in protein structure determination, e.g., CHARMM 3], ElipSys 6] and APPLAUSE 2]. Unfortunately, the running time to derive optimal solutions is still very long. For example, CHARMM takes 15 hours to nd the optimal solution of the alanine dipeptide.