{
  "version": 3,
  "created_at": "2026-07-12T04:51:17.315Z",
  "task_summary": "Mine CD4+ T-cell perturb-seq for EoE T-cell reprogramming targets: state modulation + pMHC-tolerance amenability",
  "agents": [],
  "phases": [
    {
      "name": "Plan",
      "delegations": [
        {
          "steps": [
            {
              "title": "Load DE resource + build reprogramming gene panels",
              "description": "Load GWCD4i.DE_stats.h5ad (anndata, backed). Curate and freeze four readout panels among MEASURED genes: (1) Th2 effector output (IL13,IL5,IL4,IL4R,IL17RB,PTGDR2,CCR4,CCR8); (2) Th2 master-TF/state-lock (GATA3,STAT6,IRF4,BATF,MAF,GFI1,POU2F2); (3) tolerogenic/TR1 + iTreg fate (IL10,IL21,LAG3,HAVCR2,PDCD1,TIGIT,CTLA4,ENTPD1,NT5E,FOXP3,IL2RA,IKZF2,IKZF3); (4) anergy/dysfunction (EGR2,EGR3,NR4A1,NR4A2,NR4A3,CBLB,DGKA,DGKZ,TOX,ITCH,TNFAIP3); plus a Th1 counter-axis (IFNG,TNF,TBX21,CXCR3). Emit reprogramming_panels.csv (gene, panel, measured, perturbed) and a de_qc_summary.csv (perturbations/condition, on-target KD efficiency distribution). One coverage figure."
            },
            {
              "title": "Program A — rank state-modulation targets (dampen Th2 output)",
              "description": "For each Th2-effector and master-TF readout column, extract perturbation vectors (log_fc, zscore, adj_p_value) across Rest/Stim8hr/Stim48hr. Rank perturbations whose knockdown DECREASES Th2 output (FDR<0.1), and separately those that decrease the master-TF program. Build a combined 'Th2-dampening score' (aggregating IL13+IL5+IL4 down-effects, penalizing IFNG/viability collateral). Deliverables: programA_th2_dampening_targets.csv (ranked, with per-cytokine effects + directionality), and a volcano/bar figure of top positive/negative regulators of IL13."
            },
            {
              "title": "Program B — rank pMHC-tolerance amenability co-targets",
              "description": "Score every perturbation on whether its knockdown shifts cells toward a CONVERTIBLE/tolerogenic state: up-weight increases in TR1/iTreg markers (IL10,FOXP3,IL2RA,LAG3,CTLA4) and anergy program (EGR2/3,NR4A1-3,CBLB,DGKA,TOX), and reward loss of Th2 state-lock (GATA3/STAT6/IRF4/BATF down). Define an 'amenability score' = tolerogenic/anergy induction + de-repression of regulatory fate. Cross-check against the authors' IL10/IL21 arrayed-validation table where available. Deliverables: programB_tolerance_amenability_cotargets.csv (ranked, with which tolerogenic/anergy axes each hit moves), and a figure contrasting Th2-lock effect vs tolerogenic-induction effect per candidate."
            },
            {
              "title": "Integrate polarization + robustness annotation",
              "description": "Merge in the authors' Th2_Th1_polarization_signature_DE and polarization_prediction_regulator_coefficients tables to corroborate hits with an orthogonal signature-level model. Annotate every candidate in both programs with cross-donor and cross-guide reproducibility (DE_stats.obs correlation fields) and on-target KD confidence, so weak/irreproducible hits are flagged. Deliverable: annotated master table candidates_annotated.csv and a polarization-effect vs cytokine-effect scatter."
            },
            {
              "title": "Druggability / modality classification",
              "description": "Classify each prioritized target by protein class and localization using the repo's gene-list annotations (cytokine_receptors, catalytic_receptors, gpcr, kinases, enzymes, transporters, nuclear_receptors, ion_channels) plus UniProt subcellular localization. Tag each as surface/secreted (antibody/biologic-tractable), enzyme/kinase (small-molecule), or TF/intracellular-scaffold (degrader/hard-to-drug). This sharpens Program A into biologic vs small-molecule lanes and flags which Program-B co-targets are pharmacologically reachable. Deliverable: target_druggability_classification.csv."
            },
            {
              "title": "Side-by-side condition-specificity heatmaps",
              "description": "For the top ~20 targets in each program, build heatmaps of their effect (zscore) across the full reprogramming panel and the three culture conditions, so context-specific vs constitutive regulators are visible. Two publication-quality panels: programA_heatmap.png (dampening targets × Th2 panel × condition) and programB_heatmap.png (amenability co-targets × tolerogenic/anergy panel × condition), robustness-annotated."
            },
            {
              "title": "Compile dual-program target brief for the Program Specialist",
              "description": "Write eoe_reprogramming_target_brief.md synthesizing: the explicit scope/caveats (healthy-donor in-vitro screen, hypotheses not validated EoE phenotypes, pMHC-specificity handled separately by the antigen pipeline); Program A shortlist (state-modulation targets to dampen pathogenic Th2 output, split biologic vs small-molecule); Program B shortlist (co-targets that increase amenability to pMHC-induced tolerance for anergy/iTreg approaches); the integrated evidence and robustness per target; and framed hand-off questions for the Therapeutic Program Specialist to develop both programs side by side. Embed key figures; save all artifacts."
            }
          ]
        }
      ],
      "id": "phase-0"
    }
  ],
  "feasibility": {
    "rationale": "Reframed away from the pMHC-specificity mismatch. Confirmed in the data: the Th2 effector axis (IL13/IL5/IL4/IL4R/IL17RB) and Th2 master TFs are measured+perturbed (Program A, state modulation), AND the full tolerogenic/TR1 axis (IL10/IL21/LAG3/PDCD1/CTLA4/ENTPD1), iTreg markers (FOXP3/IL2RA/IKZF2), and anergy program (EGR2/3/NR4A1-3/CBLB/DGKA/TOX) are measured (Program B, amenability to pMHC-induced tolerance). Both programs are reprogramming questions that a genome-wide regulator screen is genuinely built to answer. Outputs are target hypotheses for the Therapeutic Program Specialist, explicitly not validated EoE phenotypes.",
    "confidence": "high"
  }
}