A network update redistributes functional content and candidate priorities
Submitted to Nature Computational Science (NCS)
Abstract
Aggregate depletion capture compresses distinct changes in candidate allocation. Crossing STRING physical-network archives labelled 11.0 and 11.5 with consumed-evidence constructions in 85 colorectal models from 83 donors separated these changes. With observed BAGEL2 calls fixed, sharp finite-cohort bounds over shared unknown binary labels identified losses of 0.054–0.535 and 0.434–1.060 positive allocations per ten offered slots for PRODIGY and DawnRank. Exploratory donor reweighting retained shared core-fitness allocation contraction and oncogene-only expansion under both methods and degree ranking, whereas most within-class positive-fraction envelopes crossed zero. A same-study DawnRank comparison in 59 oesophageal models from 58 donors retained negative joint-update and shared-core allocation envelopes. Component substitutions and a score reference traced overlapping removals and opposing donor responses. Shared identities resolved contrasts and donor directions left unresolved by separate-arm bounds. Support-aware release comparison exposes allocation and identity changes compressed by an aggregate performance loss.
