rsynk/locale-benchmark-sra4571

LOCALE Benchmark: SRA-4571

Given a sequencing read, find the SRA runs whose assembled contigs it aligns to. This is the large set: 4571 SRA runs indexed, 10000 query reads, ground truth from minimap2 alignment of each read against every run's Logan contigs. Companion sets: SRA-50 v2 and SRA-500 v2, built by the same pipeline with the same filters.

Unlike the 50 and 500 sets, the runs here are not a sample: they are every run of the Metagraph 100-studies table that exists in Logan (5092), minus the 521 runs in the LOCALE training and validation sets, with no size filters. This is the first published release of the set. Embeddings are not yet published.

Runs indexed	4571
Queries	10000 (from 2145 of the runs)
Relevant runs per query	mean 17.2, median 3, max 385
Query length	43 to 301 bp, mean 138
Queries on the - strand	5339 (53%)
Filter accounting	4563409 sampled reads; 3653445 dropped for no hit to their own run, 1400 for hitting more than 915 runs, 712599 for overlapping another query; 195965 eligible, 10000 kept
Seed	5000 (final shuffle only)
Excluded	the LOCALE training and validation runs (521)

How the ground truth was built

1. Accession set. Drawn from the SRA runs of the Metagraph 100-studies table proportionally by organism, restricted to runs with at most 2M Logan contigs and at least 1 Mbp of assembled sequence, certified to exist in Logan at draw time, and with every run in the LOCALE training and validation sets removed from the pool before the draw. The draw is seeded, so accs.txt is reproducible from the seed.
2. Queries. The first 1000 spots of every run were streamed from SRA with fastq-dump (mate 1 for paired runs). Stored query sequences are the raw reads exactly as sequenced; nothing is trimmed or reverse-complemented.
3. Alignment. Every read was aligned with minimap2 (-a --eqx, both strands) against the Logan contigs of every run in the set, keeping the best alignment per (read, run) by exact-match count. An alignment counts when identity (exact matches / read length) exceeds 0.9.
4. Filters, per query, in this order: the run the read was sampled from must be in its relevant set; the read must not hit more than 20% of the runs in the set; the read must not share a contig region with another query (strict overlap filter with a 1024 bp chunk margin). Survivors are shuffled with the seed and the first N kept.
5. Mutated copies. The same queries with sequencing noise injected by mutation-simulator at SNP rate r with insertion and deletion rates r/10 each, r = 0.05 and 0.10. Every row keeps its query_id and ground-truth columns, so any method can be evaluated on identical mutated reads. The simulator's fasta and VCF per rate are under mutations/.

Files

File	Contents
accs.txt	the indexed SRA run accessions, one per line
queries.parquet	one row per query with its ground truth (schema below)
queries_mut0.00.parquet	queries.parquet plus a mutation_rate column (clean)
queries_mut0.05.parquet, queries_mut0.10.parquet	same rows in the same order with query_sequence mutated
queries.fa, queries.fa.fai	the clean queries as fasta (mutation-simulator's input) and its index
mutations/	mutation-simulator fasta and VCF for each rate

Schema of queries.parquet

Column	Type	Meaning
accession	str	run the read was sampled from
query_id	str	<accession>.<spot> from fastq-dump
query_sequence	str	the raw read
contig_accession	list[str]	relevant set: every run with an alignment above the identity threshold; always contains accession
contig_id	list[str]	best-aligned contig in each relevant run
identity	list[f64]	exact matches / read length of that alignment
ratio	list[f64]	contig length / read length
contig_len	list[i64]	contig length
aln_interval_contig	list[struct]	0-based [start, end) of the alignment on the contig
strand	str	+/-: strand of the best hit to the source run

The six list columns are parallel: element i of each describes the same alignment.

License

The data is licensed under Creative Commons cc-by-4.0; the code that built it is MIT (locale-data). The underlying sequence data comes from the NCBI Sequence Read Archive, which places no restrictions on redistribution; assembled contigs come from Logan.

Citation

For questions, reach out to ryansynk@umd.edu. If you find this work useful, please cite:

@article {Synk2026.05.12.724581,
    author = {Synk, Ryan P. and Pandey, Prashant and Sahinalp, S. Cenk and Duraiswami, Ramani},
    title = {LOCALE: Local-Alignment Embeddings for Noise-Robust DNA Search at SRA Scale},
    elocation-id = {2026.05.12.724581},
    year = {2026},
    doi = {10.64898/2026.05.12.724581},
    publisher = {Cold Spring Harbor Laboratory},
    URL = {https://www.biorxiv.org/content/early/2026/05/14/2026.05.12.724581},
    eprint = {https://www.biorxiv.org/content/early/2026/05/14/2026.05.12.724581.full.pdf},
    journal = {bioRxiv}
}