PerturbMap addresses missing perturbation responses across cellular contexts. It combines a recipient-local low-rank base with transported proposals derived from the same perturbation’s measured source responses. Source-to-recipient ridge experts are trained on paired perturbations, while proposal weights are selected using route reliability estimated from validation anchors. On a Perturb-CITE-seq melanoma cohort, the method reduces full-effect MSE by 4.1% relative to a recipient-local low-rank base. Same-recipient top-10 counterpart retrieval by cosine rises from 74.5% to 80.5%. Its MSE is within 2.82×10^-6 of a centralized pooled reference with a stronger interface.
No heat snapshots are available in the last 24 hours.