GenePerturb2Target: Directional and Context-aware Drug-target Nomination from Genome-scale CD4+ T-cell Perturb-seq

Genome-scale CRISPRi Perturb-seq resolves how each gene shapes the CD4$^+$ T-cell inflammatory program, but target-nomination pipelines rank genes by effect magnitude alone. They cannot say whether a therapy should inhibit a driver or potentiate a brake; the direction that dictates modality and human genetics is silent on it. We assign every perturbation a signed driver/brake call from the sign of its effect on a curated pro-inflammatory program across activation states. Ranking 1,923 candidates recovers known targets; the calls replicate in an independent cytokine screen ($\rho=+0.52$, IFN-$\gamma$; $\rho=+0.49$, IL-2) and correctly fail on a mismatched proliferation readout ($\rho=+0.06$). We also report that the functional axis does not out-rank genetics (AUROC 0.79 versus 0.82) but is orthogonal to it ($\rho=0.01$). The framework transfers to multiple sclerosis, rheumatoid arthritis and type 1 diabetes, where signal concentrates at the genetics$\times$direction intersection. Direction is an orthogonal, calibrated evidence axis, released as a reusable benchmark.