Borrow from networks, learn from examples, demand agreement

Settings

Watch it: Walk the network from one gene. The star map: one gene's measured links by source, widened to its neighbourhood and re-centred on a neighbour. Silent screen capture of the real application.
Strategy 12 lets every permitted network and the measurement neighbours vote on each gene's compartment, each source weighted by how far it beat chance on an inner holdout. The notebook adds a classifier (19), agreement between independent methods (31), and the understudied genes (32).
Held-out labelcompartment
Neighbours to consult15
The card strategy 12 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.
15
Console
strategy 12 test: Let every network vote, weighted by what it has earned (chance-weighted ensemble vote) -- started
layer_vote: 951 labelled genes hidden, 2,853 left to learn from
layer_vote: null 1 of 5
layer_vote: null 2 of 5
layer_vote: null 3 of 5
layer_vote: null 4 of 5
layer_vote: null 5 of 5
layer_vote: PASS
strategy 12 run: Let every network vote, weighted by what it has earned (chance-weighted ensemble vote) -- started
layer_vote: done in 1.6s
PASS: correct calls per hidden gene 0.409 against 0.102 ± 0.045 under 5 runs on shuffled labels (bar 0.143, p 0.167); effect +0.307, needed +0.050; 951 hidden
9 sources vote, each weighted by how far it beat chance on an inner holdout of the known labels; 8 earned a vote. 2,000 unlabelled genes called. Strongest: measurements (kNN) (0.28), xlms (0.12), coexpression (0.12), struct (0.07)
Hover or click a button to see what it does. Hover any setting or console line for an explanation; move the slider to step through the run.