EB Library Annotation Explorer
Choose library annotations by evidence tier, then check bile acids and carnitines against feature MS/MS. Everything runs in your browser; your files are never uploaded.
1 · Load an Everything Bagel job
Use downloaded files
Annotation TSV required. Other files enable configuration checks, coverage counts, and class screening.
2 · Selection and screening
Edit queries and read methods
One query per line: label, a tab, then the WHERE expression. The supplied skill labels its bile-acid set as 26 but lists 25; all 25 supplied queries are included. Carnitines and PEG have one default query each. Query edits are included in share links.
Every condition must match; any passing query keeps the selected annotation. Intensities use the most intense peak in tolerance relative to the spectrum base peak. Ratio references are resolved first. Equal-score ties use library name, then original row order. Catalog tiers are evidence priorities, not a numerical identification-confidence score.
Job checks
Missing libraries and catalog
Annotation coverage
Explore features
Download and share
About this app
Runs entirely in your browser: Flask only serves this page, files are never uploaded, and no CDN is used. The page is a single self-contained template with the tier catalog embedded; the line under section 2 shows its date and any gist update.
Inputs
Enter an Everything Bagel task ID and server, or drop downloaded files (library-search TSV, feature CSV, consensus MGF, submission YAML). The job loader reads /resultfile?task=…&file=… from gnps2.org or beta.gnps2.org. “Try example” uses synthetic data and needs no network.
Selection
Evidence tier: each feature takes its hit from the highest-priority tier (Tier 1 = authentic standards), then the highest cosine within it; ties fall back to library name, then row order. Selected libraries: you pick the libraries and each feature takes its best hit from them. The preset switches (compound kind, ionization mode; both come from the tier table) select every matching library in the job. Switches in a row add up and the two rows narrow each other. You can still edit the selection by hand; the page then notes that the preset was changed. Libraries are matched to the catalog by basename, case-insensitively, without the file extension; unknown libraries stay inspectable but cannot win a tier.
Screening
Bile-acid (25 queries), carnitine (1 query) and, if switched on, PEG (1 query) MS/MS checks run after selection, with every feature with a library hit is screened, whatever its name. A hit named as a class must pass that class’s queries (both, if it matches two). If it fails and its library is MassQL-derived (the *-MASSQL libraries and the GNPS C23–C28 modifications sets), it is removed and the next acceptable hit takes its place; if its library is curated, it is kept and flagged. Both behaviours can be turned off. Missing or empty spectra fail an enabled screen. Disabling a screen keeps its annotations without confirmation.
Job checks
Submitted libraries are compared with the catalog, and the library-search top-k is checked against 1000. Problems produce non-blocking warnings; discarded candidates cannot be recovered in the browser.
Downloads
Final and tier-key TSVs keep the original rows and header order; the audit JSON lists every hit, the selection and screening details, queries, source and catalog; class XLSX workbooks have Summary, Kept, Removed and Notes sheets. The final TSV is blocked if feature IDs or precursor masses disagree with the feature table.
Share links
Apply the selection first, then use “Copy share link”. Job links reload the task; local-file links restore settings but need the same files. Parameters: task, host, dataset, source (example|local), search, tier, library, screen, sort, feature, page, mode=library with libs (semicolon-separated), ordered=1, pc / pi (active compound-kind / ionization presets), bile / carnitine (1|0) and bile_queries / carnitine_queries. Links hold settings only, not spectra or results.