Skip to content

Analysis: hit reaction count is one off, moving FDR and entity ratios #296

Open

Description

@adamjohnwright

From the curator round, 19 September:

Results are now listed in the hierarchy and results in diagram are displayed in the central panel. BUT the # of hit reactions is one off (which affects the FDR and entities ratios) — see slide 5.

Why this is the most serious item in that round

Every other finding is about something being hard to see or hard to reach. This one is a number being wrong, and it propagates: the hit count feeds the FDR and the entity ratios, so a result that looks complete is quietly miscomputed. A curator comparing beta against the live site on the same analysis would see different statistics for the same data.

What is needed to start

The slide referenced in the report, or any analysis token plus the pathway where the count differs. "One off" suggests a boundary — an inclusive/exclusive range, a parent event counted with its children, or a reaction counted once per participant — but that is a guess and should not be the basis of a fix.

Reproduce against the live site side by side, since the report is a comparison rather than an absolute claim.

Raised from the curator report so it is tracked somewhere other than a Google Doc.

Activity

  1. adamjohnwright commented on Nov 13, 2026

    @adamjohnwright
    ContributorAuthor

    I tried to reproduce this on 13 Nov and couldn't. On today's data, beta and the live site agree.

    • The same release: both are on release 98 (ContentService and AnalysisService).
    • The same results: the same 20 UniProt identifiers, with and without projection to human, gave identical results on both sites for all 653 pathways found. Hit reactions found and total, entities found and total, the ratios and the FDR all match.
    • Shown as given: the new browser displays the AnalysisService's own found / total in the hierarchy and the results table, unchanged.
    • The same defaults as the live site: projection on, interactors off, disease pathways included.

    The difference seen in September may have come from beta and the live site being on different releases at the time, or the data used may hit a case my sample doesn't.

    What would let us check: the slide from that round, or the identifiers that were analysed (or the analysis token) and the pathway where the count looked one off. With those we can compare the two sites on exactly that case.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions