Work has begun exploring a relatively new class of algorithms, """"""""genetic"""""""" algorithms, to facilitate the modeling of molecular structures, specifically currently RNA molecules. These algorithms have been shown in several applications to be parallelizable and rapidly convergent to solutions in a large search space consisting of many possible suboptimal results. It is partially as a result of this research that much effort over the past year has been spent in investigating the various forms of massively parallel computer architectures to determine which offers the best performance and computational environments. This has resulted in the acquisition of an 8000 processor MasPar computer system. We are doing comparative studies with a newly developed sequence matching program based on the Smith/Waterman algorithm that runs on the MasPar. This algorithm is capable of performing 50,000,000 comparisons per second on a 4000 processor system. The results from these runs are being compared with other algorithms including another new algorithm running on a systolic array. The genetic algorithm and the MasPar architecture have been integrated into the heterogeneous system that is being developed for RNA structure analysis. This system includes the facilities for activating algorithms that may run on various computer architectures that are accessible over a computer network. The termination structures of lambda TR2 have been extensively studied and the 5' non-coding regions of Polio-virus and its mutational relationships to RNA secondary and tertiary structures are being characterized in regard to its functionality. The system has been used to study the fine structural details of the HIV-1 rev responsive element (RRE) and is currently being used to study the dimerization structure of HIV-1.
