This research aims to further develop the capabilities of a new computer simulation technique that significantly enhances the scope and versatility of usual molecular dynamics simulations. Three general directions will be explored: (i) Determination of structure of small peptides. Structure of peptides (when existent) are of significant interest. Many peptides transfer signals to receptors and investigation of their conformations may lead to the design of substitutes. Furthermore, structure of short peptides may suggest initiation site for the process of protein folding. Presence of peptide segments with high probability for a unique structure may accelerate considerably the folding process. (ii) Refinement of low resolution protein structure. Nowadays the capabilities of computational methods that are based on statistical approaches are steadily increased. Nevertheless, so far the structure obtained are of low resolution. The combination of """"""""low resolution"""""""" global search methods and atomic detail refinement procedures based on LES is a promising approach. (iii) Extension of molecular dynamics time scales. The LES provides considerably more statistics for sampling molecular events. For dynamics LES is a mean field approximation. However, with a binary collision correction developed by us the LES describes diffusion quantitatively and will be employed to study ligand escapes from a protein matrix on the tens of nanoseconds time scale, a time scale that was not accessible to molecular dynamics before.