Exploring the Impact of Pre-alignment filtering on HLS-based DNA short read alignment Accelerators

Abstract

Genomic analysis relies heavily on transforming raw sequencing data into com- plete and interpretable genomic information. At the heart of this process lies the alignment step, where billions of sequencing reads are mapped to a ref- erence genome to reconstruct the original DNA structure. Despite significant algorithmic progress, the alignment extension phase—particularly the Matrix Fill and Traceback steps—remains a major computational bottleneck due to its high time and resource demands. This thesis explores two complementary approaches to alleviate these chal- lenges: (i) the optimization and hardware acceleration of alignment algorithms, and (ii) the reduction of alignment workload through intelligent filtering. Field- Programmable Gate Arrays (FPGAs) are investigated as a hardware accelera- tion platform due to their fine-grained parallelism and configurability, which en- able efficient implementation of alignment algorithms such as Smith-Waterman, GenASM, and WFA. Furthermore, the SneakySnake pre-filtering algorithm is employed to analyze datasets, identify edit distributions, and guide algorithmic optimizations. By combining dataset-aware pre-filtering with hardware acceleration tech- niques, this work aims to minimize redundant computations, reduce the search space during alignment, and achieve substantial performance gains without compromising accuracy. The proposed system leverages insights from SneakyS- nake to dynamically adapt alignment strategies, demonstrating speedups rel- ative to software-based implementations. Experimental results and architec- tural evaluations validate the effectiveness of the proposed approach and high- light promising directions for future optimization in genomic data processing pipelines.