How much of a published figure is the catalogue? A whole-catalogue census of uncurated human microRNA entries and the literature that cites them

A 2025 paper noted that at least 1200 previously claimed human microRNAs had not been investigated. This report runs that set in full. Of 1,916 human miRBase precursors, 1,410 (73.6 per cent) have no MirGeneDB counterpart, but they carry only 12.2 per cent of the Europe PMC literature weight, so the catalogue's composition does not translate into the literature's. The set is not obscure either: every one of the 1,410 is mentioned at least once, median 62 mentions. The source paper's named examples sit at the 93.9th percentile of the set, so multiplying its average by the set size overstates the exposure roughly eightfold. Running all four annotation criteria against every entry, 570 of 1,411 fail all four and carry 51,897 mentions, while the same test fails none of the 506 curated entries. The criteria are also shown to be effectively one: of the 601 entries never reaching 1 RPM, 570 fail everything else, so the four-criterion combination adds under five percentage points to what expression alone establishes. Uncurated does not mean absent: miR-625 is detected in 312 of 320 samples and is more abundant than three quarters of curated microRNAs. Two framings of the author's own were falsified in opposite directions during the work, and both are reported.

Version 2 (18 September 2026). No measurement of the catalogue changed: the 1,917 human precursors, the 174,311 mentions carried by uncurated entries, the 20.1:1 per-entry ratio, the criterion sweep and the expression panel are all as computed on 29 August. What changed are ten statements about the source paper and about the join, each read against its source. The largest: "506 matches, zero orphans: the catalogues are perfectly nested" was wrong, because MirGeneDB's human GFF holds 514 genes of which only 506 carry a miRBase accession, and at least three of the unjoined eight sit inside this report's own "uncurated" set; the source paper counted publications with the Publish or Perish tool rather than with Google Scholar, which it never names; it examines ten cases, not five, three of them murine and two retracted years ago; and "roughly eight times too high" compared two different indexes, which run a factor of 2.0 apart on the five shared entries, making the overshoot about fourfold. The criterion sweep's denominator of 1,411 does not reconcile with the 1,410 used elsewhere and now says so. All of it is listed under "Corrections, 18 September 2026".