ABSTRACT Environmental DNA (eDNA) analysis depends critically on high‐quality reference databases. However, widely used public repositories (e. g. , NCBI) frequently suffer from annotation error, species misidentification, and sequence contamination, leading to unreliable biodiversity assessments. To address these issues, we introduce PhyloRef, a Snakemake‐based, semi‐automated phylogeny‐guided workflow for reference library curation. PhyloRef improves scalability via taxonomic grouping, detects problematic records using clustering‐based anomaly detection rather than rigid monophyly requirements, and conservatively flags ambiguous cases using a “similarₜo=” annotation. PhyloRef leverages complete mitochondrial genomes while flexibly incorporating single‐gene sequences to maximize taxonomic coverage when complete genomes are scarce. The workflow categorizes anomalies into three types: (1) single‐sequence outliers, (2) inconsistent sequence pairs, and (3) minority deviations within multi‐sequence clusters, flagging them for manual review via convenient visualizations or deleting them automatically by option. Importantly, sequences with ambiguous phylogenetic placement are annotated with a “similarₜo=” label to alert users to potential uncertainty. We validated PhyloRef using mitochondrial genome datasets for Chondrichthyes (cartilaginous fishes) and Actinopterygii (ray‐finned fishes) extracted from NCBI. The tool identified and removed nine anomalous chondrichthyan sequences and 401 Actinopterygian sequences (~2. 3% and ~5. 2% of the initial datasets, respectively), yielding curated databases of 380 sequences (266 species) and 7258 sequences (4887 species), respectively. In addition, nine sequences were flagged with “similarₜo=” label in chondrichthyan fishes and 597 in Actinopterygian fishes, to reduce the risk of misidentification in downstream eDNA analyses. This resource enhances the reliability of eDNA‐based biodiversity and ecological studies. Future directions include integrating machine learning for anomaly detection, incorporating nuclear markers for improved taxonomic resolution, and developing automated updating modules.
Mai et al. (Thu,) studied this question.