Fragmented mobile genetic element benchmark: plasmids and phage-plasmids

Overview This archive provides a fragmented sequence benchmark for evaluating detection and recovery tools on mobile genetic elements. Reference genomes for two element classes (plasmids and phage-plasmids) were cut into non-overlapping fragments of fixed length, and randomised subsets were drawn to serve as mock datasets. Every fragment is traceable to its source sequence and carries a class label, so the archive can be used directly as labelled ground truth for classification, detection, or recovery tasks. Archive contents The archive contains two zipped folders, one per element class: plasmid_mocks.zip phageplasmid_mocks.zip Within each folder, files are organised by fragment length (500, 1000, 3000, and 5000 bp). For each length there are five subset FASTA files and one manifest: {class}{length}bp__subset_001.fasta ... {class}{length}bp__subset_005.fasta {class}_{length}bp__manifest.tsv where {class} is plasmid or phageplasmid and {length} is one of 500, 1000, 3000, 5000. Each subset FASTA contains 10,000 fragments. With five subsets per length and four lengths per class, each folder holds 20 FASTA files and 4 manifest files (50,000 fragment records per length per class, across the five subsets). Data provenance Plasmids: complete plasmid sequences from PLSDB, version 2024_05_31_v2 (https://ccb-microbe.cs.uni-saarland.de/plsdb2025/, accessed August 2026). PLSDB sequences are already dereplicated at the record level, so no additional deduplication was applied. Phage-plasmids: sequences retrieved by NCBI nucleotide accession from the phage-plasmid set of Pfeifer and Rocha (2024); see the citation below. How the fragments were generated Fragmentation: each source sequence was cut into non-overlapping fragments of a single fixed length, with one pass per target length (500, 1000, 3000, 5000 bp), using default options. Fragments are exact length: any trailing sequence shorter than the target length was discarded, and source sequences shorter than the target length produced no fragment. Sequences were treated linearly, so circular plasmids were not wrapped across the origin and no fragment spans the sequence breakpoint. Subsetting: for each class and length, five subsets of 10,000 fragments were drawn from the pooled fragments using a fixed random seed of 42. Subsets were drawn independently from the shared pool (default, non-disjoint mode), so a given fragment may appear in more than one of the five subsets. The fixed seed makes the draw fully reproducible. File formats FASTA fragment headers encode both provenance and ground truth. Each header has the form: >SOURCE_fragNNNNNN source=SOURCE start=S end=E fraglen=L label=CLASS where SOURCE is the source sequence identifier, NNNNNN is a zero-padded fragment index within that source, S and E are the 1-based inclusive start and end coordinates of the fragment within the source, L is the fragment length in bp, and CLASS is the element class label (plasmid or phageplasmid). Manifest files are tab-separated and map every fragment in the five subsets of a given length to its origin. Columns: subset: subset identifier fragment_id: unique fragment identifier (matches the FASTA header identifier) source: source sequence identifier the fragment was cut from label: element class label start: 1-based start coordinate within the source end: 1-based end coordinate within the source fraglen: fragment length in bp The manifest is the recommended ground-truth reference for scoring: join a tool's predictions to it on fragment_id to evaluate by element class and by fragment length. Suggested use Run the tool under evaluation on the subset FASTA files, then join the predictions to the corresponding manifest on fragment_id to score performance by class and by fragment length. The four lengths are intended to characterise how detection or recovery degrades on shorter fragments. To avoid circular evaluation, keep this reference set independent of any tool whose own reference database overlaps these source sequences. Reproducibility The fragments and subsets were produced with two scripts (a fixed-length chopper and a random subset sampler). Re-running them with the same input references, fragment lengths, and random seed (42) reproduces the subsets exactly. Scripts are provided in this archive. How to cite If you use this dataset, please acknowlege this Zenodo (10.5281/zenodo.21900396) Please also cite the underlying data sources: PLSDB (version 2024_05_31_v2). https://ccb-microbe.cs.uni-saarland.de/plsdb2025/, accessed August 2026. Pfeifer, E., Rocha, E.P.C. Phage-plasmids promote recombination and emergence of phages and plasmids. Nat Commun 15, 1545 (2024). https://doi.org/10.1038/s41467-024-45757-3