Abstract
BACKGROUND: Shotgun metagenomics enables to study microbial communities without biases from culturing and isolation, but taxonomic classification to the species level remains challenging due to high false positive rates. Oxford Nanopore Technologies offers new opportunities to address these challenges by producing longer reads. However, different pipelines and tools use different methods to reduce false positives, resulting in variable outcomes with limited exploration of what works best in practice. Relative abundance filtering is often used to improve precision by removing false positives but reduces also recall by removing true positives. In this study, we optimized a broadly applicable taxonomic classification pipeline for long-read nanopore sequencing data that improves precision. The pipeline uses the tool KMA as the underlying classifier, followed by specific post-processing and optimization of filtering thresholds. Based on ten defined mock communities, different filter thresholds were evaluated, alongside the effect of the sequencing yield and the limit of detection (LOD). RESULTS: Our optimized pipeline substantially outperformed default classifier settings, and the conventionally used relative abundance filtering. Classification accuracy improved with higher sequencing yields, requiring at least a post-filtering yield of 500M bases, and ideally 1000M bases, for reliable results. At yields above 1000M bases, median precision could be improved up to 95% while maintaining median recall at 91.62%. Further increasing median precision to 99% reduced recall to 79.08%. Similarly, higher sequencing yields lowered LOD. For yields above 1000 M bases, the limit of detection remained stable at 0.1% up to a median precision of 95%, while yields below 1000M showed an LOD of 1%. Validation on ten probiotic-derived mock communities confirmed the pipeline’s performance and general applicability. CONCLUSION: Our optimized classification pipeline for nanopore sequencing data provides substantially higher precision compared to default approaches and is suitable for diverse metagenomic applications. We provide specific guidance on expected recall and precision values for minimum sequencing yields and their associated LODs. Our optimized pipeline, called KAPTAIN (KMA-bAsed Pipeline for meTAgenomic specIes ideNtification), is publicly available on GitHub (https://github.com/BioinformaticsPlatformWIV-ISP/KAPTAIN) and also the Galaxy instance of our institute (https://galaxy.sciensano.be) to be used by other scientists. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12864-026-12668-0.