neuralbioinfo/enterobacterales-plasmid-training

Enterobacterales Clean-v2 Segment Dataset

This dataset contains the Enterobacterales clean-v2 NR plasmid training segments and their sampled host candidate segments. Problematic host segments are not dropped from the canonical export; they are retained and tagged.

Configs

config	split	rows	has_sequence
enterobacterales_clean_v2_host_assemblies	train	19780	False
enterobacterales_clean_v2_host_contigs	train	380200	False
enterobacterales_clean_v2_plasmid_to_assembly_links	train	100037	False
enterobacterales_clean_v2_plasmids	train	29590	False
enterobacterales_clean_v2_segments	train	9740250	True

Only enterobacterales_clean_v2_segments contains the nucleotide segment sequence column.

Counts

metric	value
plasmid segment rows	1948050
host candidate rows	7792200
total canonical rows	9740250
host rows flagged by plasmid overlap	1851270
current clean-training eligible rows	7888980
unique source sequence IDs	165321
segment Parquet shards	98
target compressed shard size	100 MB
selected Enterobacterales clean-v2 NR plasmids	29590
host assemblies with usable non-plasmid contigs	19656

Segment Flags

• segment_id: unique integer primary key, contiguous from 0.
• sequence_id: source-sequence grouping ID, contiguous from 0; segments from the same plasmid or host source contig share one value.
• labels: 1 for plasmid segments, 0 for host segments.
• training_eligible_current_clean: current recommended training filter.
• filtered_by_plasmid: host segment has a likely plasmid-overlap minimap2 hit.
• is_problematic_segment: currently equivalent to a host segment flagged by plasmid overlap.
• problematic_reasons: text reason for problematic rows, currently host_plasmid_overlap.

Loading

from datasets import load_dataset

segments = load_dataset("neuralbioinfo/enterobacterales-plasmid-training", "enterobacterales_clean_v2_segments", split="train")
clean = segments.filter(lambda row: row["training_eligible_current_clean"])
flagged_hosts = segments.filter(lambda row: row["source_type"] == "host" and row["filtered_by_plasmid"])

Provenance

• Source clean-v2 plasmid DB: clean_v2_len2k_4mb_nochrom_broadhq20
• Taxonomic scope: Enterobacterales order taxid 91347
• Segment length: 4000
• Plasmid coverage: 2.0
• Host candidate ratio: 4x
• Host-overlap filter: minimap2 against selected Enterobacterales clean-v2 NR plasmids.

Detailed JSON reports are stored under reports/.