Umi-pipeline-nf: a modular and scalable workflow for UMI-tagged nanopore amplicon analysis with real-time sequencing integration and GPU-acceleration

<h4>Motivation</h4>Unique molecular identifiers (UMIs) enable efficient error correction in amplicon sequencing but UMI-aware analysis workflows for long-read sequencing and particularly for nanopore data are still sparse. Existing approaches lack portability, real-time sequencing support, GPU acceleration, and efficient use of resources.<h4>Results</h4>We present umi-pipeline-nf, a portable, fully containerized, modular and scalable workflow to create single-molecule consensus sequences from UMI-tagged long-read nanopore amplicon data. Umi-pipeline-nf supports flexible UMI-designs and is built in Nextflow DSL2 for seamless deployment across computing platforms and a high degree of parallelization, allowing analysis of several targets at once. It scales linearly from single samples to large cohorts, outperforming existing tools in efficiency and flexibility. Additionally, we integrated real-time read processing, robust UMI clustering, and GPU-accelerated consensus polishing. Umi-pipeline-nf supports two different polishing strategies [reference sequence-based and partial order alignment (POA)-based]. Implementation of GPU-accelerated, reference sequence-based polishing results in up to 100-fold runtime improvements and reduced usage of computational resources, compared to other UMI analysis pipelines and POA-based polishing.<h4>Availability and implementation</h4>The umi-pipeline-nf analysis pipeline and test data are available at https://github.com/genepi/umi-pipeline-nf, and a frozen snapshot is available at DOI: 10.5281/zenodo.18607956. Scripts and configuration files for the analyses in the present manuscript can be found at https://github.com/AmstlerStephan/umi-pipeline-nf_Paper.