Core Concepts
AGAThA proposes an efficient GPU-based acceleration method for guided sequence alignment, achieving significant speedups compared to existing baselines.
Abstract
AGAThA addresses the challenges of aligning long reads by proposing novel strategies like rolling window, sliced diagonal, subwarp rejoining, and uneven bucketing. Experimental results show AGAThA outperforms existing baselines significantly across various datasets.
AGAThA's performance is compared against Manymap, GASAL2, SALoBa, and LOGAN. AGAThA consistently shows superior speedups over these baselines in aligning real-world sequencing datasets.
Stats
AGAThA achieves 18.8× speedup against the CPU-based baseline.
AGAThA outperforms Manymap by 12.1× and GASAL2 by 36.6×.
AGAThA shows a 3.6× speedup over SALoBa in GPU-accelerated sequence alignment.
AGAThA surpasses LOGAN's performance closely following SALoBa.