Heterogeneous Hardware-assisted Parallel Processing for BWA-MEM DNA Alignment

With the fast development in technology, the amount of genetic data needs to be processed grows exponentially. As an evidence for this fact, there have been a huge number of modern tools/approaches published over the last decades. DNA alignment is one of the most important stages in biological investigations, and it also attracts much attention. Although there exists many algorithms that significantly improve processing time of this stage, it still takes up to a number of days for aligning the whole human genome even with large clusters. In this work, we accelerate the improved BWA-MEM algorithm, an efficient DNA alignment algorithm, with our proposed architecture. The implementation discussed in this work focuses on exploitation of the pipeline architecture and optimization in memory access as well as data communication. Our system can obtain speed-ups by up to 14.3× when compared with the other computing platforms.