Abstract

The strand separation problem is that of calculating the probability of separation of each base pair in the DNA sequence under specified physical and chemical conditions. A computational method that performs an approximate statistical mechanical analysis of this phenomenon has previously been developed and shown to be accurate when compared with experiments. In this paper we propose a parallelization for the DNA strand separation prediction problem and evaluate the performance of two parallel implementations on a shared memory multiprocessor, one using static and the other dynamic scheduling. We show that DNA strand separation can be efficiently parallelized using a loop-based decomposition. For the DNA molecule of about 4K bases we chose as input, static scheduling performs well on up to 8 processors, but causes substantial load imbalance on 16 processors. Dynamic scheduling delivers a very good speedup in all considered cases. Keywords: DNA Strand Separation, Parallel Applications, Dynamic Scheduling 3 Supported by grants RO1-GM47012 (NIH) and BIR 93-10252 (NSF). y Supported by NSF Career Development Award CCR-9501942, an Alfred P. Sloan Research Fellowship and NATO Grant 960215. 1

Citations