/target-prioritization

Per-gene rationale template

When filling the Rationale and Suggested next step slots in targets_report.md, follow this template per gene. Keep it tight — 2-3 sentences total for rationale, 1 sentence for next step.

Rationale (2-3 sentences)

Sentence 1 — most compelling evidence (pick the single strongest signal from the dossier row):

• If is_focus_disease_associated → "OpenTargets surfaces a focus-disease association (score [max_focus_disease_assoc_score]) — likely backed by GWAS / text-mining evidence…"
• Else if any_focus_disease_drug → "Already targeted by an approved focus-disease drug ([drug name])…"
• Else if depmap_pct_essential >= 0.3 AND depmap_pct_essential <= 0.8 → "Selectively essential in [N]% of DepMap cell lines (mean geneEffect [val]) — clear dependency in the relevant lineage…"
• Else if chembl_best_pchembl >= 9 → "ChEMBL surfaces a sub-nM tool compound ([compound name], pIC50 [val]) — chemistry available now for ex-vivo validation…"
• Else if hpa_focus_cell_hits non-empty AND cell_context_score >= 0.7 → "HPA single-cell data ranks [cell type] in the top-2 expressing cell types for this gene (nCPM [val]) — strong target-cell expression…"
• Else if tissue_specificity == 1.0 (HPA Tissue enriched / Group enriched) → "Narrow tissue expression in [top tissue] (HPA enriched) — cleaner therapeutic window than a broadly expressed target…"
• Else if is_surface AND highest_clinical_phase >= 3 → "Surface protein with phase III drugs in adjacent indications…"
• Else if is_surface AND maturity_tag in {novel, moderate} → "Surface protein with moderate prior literature ([N] PubMed hits) — tractable for antibody / CAR / ADC approaches…"
• Else use whatever component scores highest in the breakdown row.

Sentence 2 — main risk or caveat:

• maturity_tag = uncharted → "Very thin literature ([N] hits) — risk of unknown off-target biology."
• maturity_tag = saturated → "Heavily studied ([N] hits); likely IP crowded."
• is_mhc = True → "MHC-family gene — broad-spectrum effects, hard to inhibit selectively."
• composite driven by single dimension only → "Score concentrated in one dimension — confirm with orthogonal evidence before pursuing."
• No disease association → "Stat signal from DE only; lacks genetic corroboration."
• tissue_specificity <= 0.2 (HPA Low tissue specificity) → "Broadly expressed across tissues — narrow therapeutic window unlikely without a delivery / targeting strategy."
• cell_context_score == 0 and hpa_focus_cell_hits empty → "Not in the top-expressing cell types per HPA single-cell data — efficacy in the target population uncertain."
• depmap_pct_essential >= 0.85 → "Pan-essential in DepMap (≥85% of cell lines depend on it) — broad cytotoxicity risk; therapeutic window unlikely without selective delivery."
• safety_constraint_score <= 0.4 (LOEUF in top decile) → "gnomAD flags this gene as highly LoF-constrained (LOEUF [val]) — full inhibition may approach haploinsufficient territory."
• chembl_target_id is None or chembl_best_pchembl is None → "No potent IC50 tool compound in ChEMBL (pIC50 ≥ 7) — chemical biology starting point is limited."

Sentence 3 (optional) — specific project context if obvious from the dossier (e.g. "Persists in non-responder cells at post-treatment, consistent with [pathway] escape"; "Co-expressed with [marker] in the tumour microenvironment dataset"; "Up in [region] but absent from healthy control biopsies").

Suggested next step (1 sentence)

Be concrete and adapted to the user's experimental context. Examples across domains:

• Functional genomics: "siRNA / CRISPRi knockdown in the relevant primary cell type; readout the disease-relevant secreted protein or phosphorylation marker by ELISA / flow."
• Histology: "IHC / RNAscope on patient vs healthy tissue from the matching anatomical site to confirm protein-level upregulation."
• Pharmacology: "Treat ex-vivo with [tool compound from approved_drugs] and compare against the standard-of-care arm."
• Replication: "Cross-check expression in an independent public cohort with matched contrast (e.g. a GEO / ArrayExpress dataset)."
• Chemistry: "If no tool compound exists: structure-based virtual screen against UniProt:[id], or commission a fragment-screen pilot."

Executive summary (top 5–10 genes)

3–5 sentences total at the top of the report. Cover:

1. How many genes scored Tier-1 vs Tier-2.
2. The 2–3 most compelling individual candidates and the headline reason.
3. Any pattern across the top genes (e.g. "5/10 are surface receptors in the same signalling cassette, suggesting [pathway] is the dominant axis").
4. Caveats / what's missing (e.g. "All candidates derive from a single contrast — recommend cross-checking against an orthogonal one").

Keep it factual. Do not invent biology not supported by the dossier rows or the original DE context.