Scalable Accelerator Architecture for Local Alignment of DNA Sequences

The Smith-Waterman algorithm is widely used to determine the optimal sequence alignment between two DNA sequences. This paper presents an innovative method to significantly reduce the computation time and memory space requirements of the traceback phase of this alignment algorithm. It also presents a flexible and scalable hardware architecture for accelerating such method, which can be easily expandable by the interconnection of several FPGAs. The results obtained from an implementation using a Virtex-5 FPGA showed that the proposed method is highly feasible in order to provide significant gains in terms of the overall performance of the whole alignment procedure when long sequences are processed. The obtained results also showed that it is preferable to span the array of processing elements through several FPGAs, rather than reusing the hardware resources of the individual array.