# Overlap Desk > Paste two genomic region sets as BED - peaks and promoters, two replicates, two conditions - and > find out what their overlap supports. The browser computes the overlap exactly as gtars 0.9.2 > does, compares it with chance by a seeded within-chromosome shuffle and, given a universe, a > Fisher's exact test, and flags inputs that cannot be compared. Then a metered reading explains > what the overlap supports, or writes the gtars script that reproduces every number. URL: https://overlap-desk.skillsafe.ai/ API: https://overlap-desk.skillsafe.ai/api.html Model: gpt-terra · publisher markup 1000 bps (10%) Not a clinical or diagnostic tool. ## The free engine (in the browser, no account) - Reads BED as gtars does: 0-based half-open, browser/track/# lines skipped, a first line with a non-numeric start read as a header, at least three columns, start < end. Invalid lines are listed and left out. Up to 200,000 intervals per set and 300,000 universe regions. - Metrics (gtars semantics; intervals overlap when they share a base, book-ended ones do not; overlapping and book-ended intervals merge before base-pair metrics): Jaccard, coverage of A by B and of B by A, overlap coefficient, intervals of each set with an overlap, per-interval overlap counts, the gap to the nearest interval. - Chance: expected shared bases under uniform placement per chromosome; a shuffle test that moves each B interval uniformly within its own chromosome, keeping its width (bedtools shuffle -chrom), up to 1,000 shuffles with a fixed seed; with a universe, a one- and two-sided Fisher's exact test on universe regions that overlap A, B, both or neither (the LOLA design). - Genomes: hg38, hg19, mm10, mm39 primary chromosomes from UCSC chrom.sizes, or a pasted chrom.sizes. Contig names are compared exactly. - Flags: mismatched contig naming (chr1 against 1), no shared contigs, intervals past a chromosome end with the assembly they do fit, unknown contigs, invalid lines, self-overlapping and duplicated intervals, thin overlap, widths that differ 20-fold, no background, a whole-genome background for peak-like sets, sets outside the universe, few shuffles, strand ignored, and a set cut at the row limit. Contig naming and wrong-assembly flags carry a one-click fix (rename chr1 <-> 1, chrM <-> MT, or switch to the assembly every interval fits). - Checked against gtars 0.9.2, bedtools 2.26.0 and scipy 1.15.3 on 300 random cases (38,679 comparisons, no disagreement). - Exports: a.bed / b.bed / universe.bed (the valid intervals, tab-separated), A annotated with overlap counts and gaps (CSV), per-chromosome table (CSV), summary (Markdown) with a methods paragraph, and a_hit.bed / a_nohit.bed / b_hit.bed (intervals with or without an overlap). A file made from a set cut at the row limit says so in its name (.partial.bed, -partial). ## The metered lanes (input field `task`) - `interpret` - a reading of each metric, each chance test with the browser's result, the user's claims judged against the facts, and what the overlap cannot show. Verdict sound / caveated / unreliable, never looser than the browser's read unless its flags are dismissed. - `script` - a Python script that loads a.bed and b.bed with gtars RegionSet, checks every expected value with math.isclose, runs scipy.stats.fisher_exact with a universe, and applies the fixes (renaming contigs, dropping bad intervals, merging, a universe test, a seeded shuffle). Every reply is reconciled on the page: every flag answered, every number found in the browser's facts or the user's notes, test results copied, script paths and expected values checked. Only the analysis's statistics are sent, never the intervals. ## Sources - Derived from the agent skill @k-dense-ai/gtars (https://skillsafe.ai/skill/@k-dense-ai/gtars), k-dense-ai/scientific-agent-skills by K-Dense Inc. (MIT). - gtars: https://github.com/databio/gtars · LOLA: Sheffield & Bock, Bioinformatics 32(4) 587-589, 2016, doi:10.1093/bioinformatics/btv612. - Notice: https://overlap-desk.skillsafe.ai/NOTICE.txt