Asensitive Compression-Based Method for Filtering Targeted FASTQ Sequencing Reads

Bioinformatic $s$ applications often necessitate filtering FASTQ sequencing reads based on their similarity to specific sequence targets, such as eliminating contamination or isolating reads associated with particular viruses. While alignment-based methods are effective for these tasks, they exhibit reduced sensitivity and may introduce overestimations, especially when confronted with lower similarity searches. In this paper, we in-troduce a novel alignment-free method for filtering FASTQ reads according to defined similarity thresholds. Unlike alignment-based approaches, our method maintains higher sensitivity even in scenarios with lower similarity targets, for example, in ancient DNA. Additionally, our approach, being compression-based, mitigates the risk of overestimation inherent in other methods. We demonstrate the versatility of our method across various applications and provide a public, open-source imple-mentation called MAGNET. MAGNET offers multi-threading capabilities for accelerated processing and is freely accessible at https://github.com/cobilab/magnet.