Settings
Watch it: Pick a map that shows your label. The pregenerated gallery, sorted by how well a label maps onto each, and the score table that says so. Silent screen capture of the real application.
Held-out labelcompartment
UMAP neighbours to try15, 50
UMAP min_dist to try0.0
Cluster sizes to try20, 50
Cluster selection to tryeom, leaf
Genes per map2000
Feature setsall
Call a label's cluster at F10.2
The card strategy 01 opens on: the same four bars as every strategy (better than chance, reach, two plain task metrics, each with its interval and chance), Run / Test / Details ▸, what its test does, and one real failure beside one real success.
Console
strategy 01 test: Hold out a category and search for a map that finds it (UMAP + HDBSCAN) -- started
map 1 of 2: all permitted, n_neighbors 15, min_dist 0.0
map 2 of 2: all permitted, n_neighbors 50, min_dist 0.0
holdout_search: PASS
strategy 01 run: Hold out a category and search for a map that finds it (UMAP + HDBSCAN) -- started
map 1 of 2: all permitted, n_neighbors 15, min_dist 0.0
map 2 of 2: all permitted, n_neighbors 50, min_dist 0.0
holdout_search: done in 0.2s
PASS: F1 of hidden genes in the cluster chosen for their label on known genes 0.106 against 0.033 ± 0.009 under 100 permutations of the hidden genes' labels over the same chosen clusters (bar 0.048, p 0.0099); effect +0.073, needed +0.050; 426 hidden
Best of 8 configurations: all permitted, n_neighbors 15, min_dist 0.0, min cluster 20 (leaf). Mean F1 0.271 over the compartment labels, against 0.160 ± 0.006 for the same clustering with the labels shuffled; 2 label(s) recovered at F1 >= 0.5, best 60S ribosome at 0.58. 37 unlabelled genes sit in the cluster that best isolates a label at F1 >= 0.20, and are called with it; 'support' is that F1.