Target Validation & Tractability

Decide go/no-go on whether a gene is a causal, druggable disease target. The agent has read-only tools over the standard validation stack — Open Targets association + evidence datatypes, DepMap (Project Achilles) CRISPR gene-effect with selective-vs-pan-essential, Open Targets/GSK tractability buckets, and human genetic evidence (GWAS/OMIM) + mouse-KO (IMPC) concordance. Tools return raw scores; the agent must integrate genetics, dependency interpretation, and tractability into a GO / CONDITIONAL / NO-GO call with calibrated confidence — the single highest-leverage decision in drug R&D, since most pipeline failures are wrong-target failures.

Why this is fundable

Agent tools

Expert grading rubric

Dimension	5 (excellent)	1 (poor)
Genetic / causal evidence weighting	Treats human genetic evidence (tier, OT genetic_association, KO concordance) as the dominant validation axis, and weights clear causal (Mendelian/coding) genetics accordingly; notes direction-of-effect (e.g. LoF risk allele needs an agonist).	Ignores or under-weights genetics, conflates somatic-driver with germline causal evidence, or treats a high overall OT score as validation without checking which datatype drives it.
Dependency interpretation (selective vs pan-essential)	Correctly reads the DepMap gene-effect: recognizes a selective dependency as a plus and a pan-essential / common-essential gene as a toxicity red flag (no therapeutic window), not as evidence the target is 'important'.	Treats any strongly negative gene-effect as a good thing, misses the pan-essential trap (MYC/POLR2A), or never distinguishes selective from common-essential.
Tractability / modality fit	Maps the target to a viable modality (small molecule vs antibody) using the tractability buckets and target biology (e.g. secreted cytokine -> antibody; kinase -> small molecule; intrinsically disordered TF -> low tractability both ways).	Recommends a modality the tractability data contradicts, ignores tractability entirely, or asserts druggability with no structural/precedence basis.
Integrated go/no-go judgment & calibration	Synthesizes genetics + dependency + tractability into a clear GO / CONDITIONAL / NO-GO that matches the evidence, with calibrated confidence and the right caveats (e.g. LRRK2 strong genetics but failed clinical readout -> conditional, not clean GO).	No clear verdict, or a verdict contradicted by its own evidence; overconfident on ambiguous cases or hedges on clear-cut ones.
Evidence faithfulness	Every score and claim (association, gene-effect, tractability, genetic tier) traces to the tool outputs; no fabricated numbers or invented variants.	Hallucinates scores, variants, or programs, or contradicts the returned tool data.

Example queries

• Validate LRRK2 as a therapeutic target for Parkinson's disease — is the causal disease link and druggability strong enough to advance a small-molecule program?
• Assess PCSK9 as a target for hypercholesterolemia / cardiovascular risk. Integrate the genetic, dependency, and tractability evidence into a go/no-go recommendation.
• A team proposes MYC as a target for solid tumors. Evaluate whether this is a tractable, well-validated target or a trap, and give a go/no-go with confidence.
• We see a very strong DepMap dependency score for POLR2A and want to pursue it as a cancer target. Validate this target — is the dependency a real selective vulnerability or a pan-essential artifact?

Trajectories

model panel (compare side by side)

Model	Provider	Tier	Judge 1–5	Verdict
Claude Opus 4.8	anthropic	frontier	4.6	acceptable
GPT-4o mini	openai	small	4.2	acceptable
GPT (frontier)	openai	frontier	4.2	acceptable
Claude Haiku 4.5	anthropic	small	4.0	acceptable