21 indications in our corpus have a disease expression signature we hold but cannot score reproducibly. This page states, for each, exactly what we have and how far short it falls. It is published because it is the part of this work most likely to be useful to someone attempting the same thing, and because the alternative — a ranked list we cannot defend — is worse than nothing.
Our score is a rank correlation between a disease signature and a compound's LINCS profile, computed only over the genes the two share. Every signature below carries roughly 600 genes — they are not small. They are measured on a different gene space: only 9 to 78 of those genes fall inside the 978-gene LINCS landmark set, and that intersection is all the score has to work with.
We measured how far that can be pushed. Across six deep signatures, subsampled and re-scored twenty times each against a 200-permutation null, a ranking reproduced in 99.2% of draws at 200 shared genes, 91.7% at 150 and 80.8% at 100. At 50 the result could not be separated from the same signature with its gene labels shuffled. Our floor is 300 — above the deepest point we actually measured, deliberately. Everything in this table sits below 80.
What this does not say. It does not say the method works above the floor. That is a separate question, and the only control it has passed is a specificity check: scrambling a signature's values collapses the candidate list to zero overlap with the real one. Cross-signature replication remains untested. See method & limits.
| Indication | Signature | Genes | In landmark set | Corpus percentile | Reproduces at this depth |
|---|---|---|---|---|---|
| breast cancer | CREEDS dz:392 (GSE1379, human) | 600 | 9 | 0th | indistinguishable from shuffled |
| type 2 diabetes mellitus | CREEDS dz:895 (GSE23343, human) | 600 | 13 | 1th | indistinguishable from shuffled |
| melanoma | CREEDS dz:950 (GSE6887, human) | 600 | 30 | 2th | indistinguishable from shuffled |
| swine influenza | CREEDS dz:498 (GSE48466, human) | 597 | 42 | 8th | indistinguishable from shuffled |
| liver cirrhosis | CREEDS dz:1012 (GSE50892, human) | 577 | 43 | 11th | indistinguishable from shuffled |
| acute T cell leukemia | CREEDS dz:351 (GSE10789, human) | 600 | 44 | 12th | indistinguishable from shuffled |
| glioblastoma multiforme | CREEDS dz:862 (GSE15824, human) | 598 | 48 | 16th | indistinguishable from shuffled |
| Duchenne muscular dystrophy | CREEDS dz:674 (GSE6011, human) | 595 | 49 | 17th | indistinguishable from shuffled |
| NASH | CREEDS dz:185 (GSE24807, human) | 596 | 51 | 19th | not measured |
| epilepsy syndrome | CREEDS dz:417 (GSE7486, human) | 584 | 52 | 20th | not measured |
| acute myocardial infarction | CREEDS dz:510 (GSE48060, human) | 596 | 53 | 21th | not measured |
| Schizophrenia | CREEDS dz:226 (GSE12654, human) | 600 | 59 | 25th | not measured |
| osteoporosis | CREEDS dz:163 (GSE2208, human) | 596 | 59 | 25th | not measured |
| HIV encephalitis | CREEDS dz:2 (GSE3489, human) | 599 | 61 | 27th | not measured |
| Diabetic Nephropathy | CREEDS dz:223 (GSE1009, human) | 599 | 61 | 27th | not measured |
| Senescence | CREEDS dz:408 (GSE1572, human) | 599 | 63 | 32th | not measured |
| colorectal cancer | CREEDS dz:552 (GSE32323, human) | 599 | 64 | 33th | not measured |
| pancreatic cancer | CREEDS dz:483 (GSE18670, human) | 598 | 65 | 34th | not measured |
| mental depression | CREEDS dz:462 (GSE12654, human) | 598 | 67 | 35th | not measured |
| renal cell carcinoma | CREEDS dz:664 (GSE38424, human) | 600 | 73 | 36th | not measured |
| pulmonary hypertension | CREEDS dz:241 (GSE703, human) | 600 | 78 | 38th | not measured |
Each indication links to its own page, which lists the specific landmark genes that signature covers and the compounds it surfaced — alphabetical and deliberately unranked, with our structural-toxicity flags and registered-trial counts shown rather than hidden. Those pages are not indexed: on their own each is this table's row plus a compound list, and a compound list is not a reason to rank in search.
A signature for the indication measured on a platform overlapping the landmark set more deeply. We cannot tell you which genes would matter most for a given disease — that needs a curated disease-gene map we do not hold and are not going to assert. If you have a deeper signature and want it scored against our library, that is a conversation we are glad to have.