6 signatures in our corpus were measured in mouse or rat. To score them against a human compound library their gene symbols had been converted to human by making them upper case — Trp53 becomes TRP53. That is a string operation, not an orthology assignment, and we do not ship a ranking built on one.
Two of these indications are still sold — sepsis and lung cancer — because we also hold a deep human signature for each, and those pages are built from it. The mouse signature below is not used to produce them. We list it anyway: we hold it, and a page about a mapping error that quietly omitted two of its own cases would be the same failure in a different place.
A wrong answer that looks wrong is cheap. This one produces answers that look right.
Our progeria page rested on a 57-gene mouse signature. Among the compounds it surfaced, pravastatin appeared second. Pravastatin is not a random compound in progeria: it has been through three registered Hutchinson-Gilford trials as part of the lonafarnib–zoledronate–pravastatin combination. A reader arriving at that page would have seen the method recover a known progeria therapy and reasonably concluded it works.
It did not. The signature is mouse, the overlap is 57 of 978, and the symbols were uppercased. The agreement is a coincidence, and it is the most dangerous kind — one that lands on the page where we have the least evidence, and that the honest caveat above it makes more credible rather than less. We have removed the compound names from all of these pages. Naming the organism and the mapping error is the transparency that matters here; the compound names add nothing a reader can use and carry the entire risk of false confirmation.
| Indication | Organism | Landmark overlap | Signature | What happened to the page |
|---|---|---|---|---|
| fatty liver disease | rat | 39 | CREEDS dz:68 (GSE665, rat) | withheld |
| Sepsis | mouse | 39 | CREEDS dz:150 (GSE4479, mouse) | sold on a separate deep human signature; this mouse ranking is not used |
| lung cancer | mouse | 41 | CREEDS dz:615 (GSE14277, mouse) | sold on a separate deep human signature; this mouse ranking is not used |
| congestive heart failure | mouse | 47 | CREEDS dz:273 (GSE1988, mouse) | withheld |
| progeria | mouse | 57 | CREEDS dz:984 (GSE32609, mouse) | withheld |
| Kidney disorder associated with type 2 diabetes mellitus | mouse | 62 | CREEDS dz:8 (GSE2557, mouse) | withheld |
Species is checked before gene depth, and blocks separately. That ordering is deliberate: all of these also fall below our depth floor, so a depth-first check would have attributed every one to a gene count and the species problem would never have appeared in an audit — until a deep mouse signature arrived, cleared the depth gate and shipped labelled reliable. The failure a gate hides is the failure it will eventually let through.
A real ortholog mapping (HomoloGene, Ensembl Compara, or MGI homology classes) with the unmapped fraction reported is the minimum bar for us to score a cross-species signature at all. We have not done that work, so no ranking on this site is produced from a cross-species signature.