Improving the performance of Smith-Waterman sequence algorithm on GPU using shared memory for biological protein sequences

In Bioinformatics, sequence alignment algorithm aims to find out whether biological sequences (e.g., DNA, RNA, or Protein sequences) are related or not. A variety of algorithms are developed, Smith-Waterman Algorithm (SW) is a well-known local alignment algorithm to find the similarity of two sequences and provides optimal result using dynamic programming. As the size of sequence database is doubling about every 6 months, the computational time also increases. Sequence alignment algorithms performance can have improved by using the parallel computing technology on the GPU. In this paper, we proposed a method to improve the performance of SW algorithm by using GPU's shared memory instead of global memory. By using shared memory, the data being transferred between the global memory and processing elements is reduced, which in turn improves the performance. The tabulated result showed positive sign of correctness in proposed method and tested using UniProt sequence database.