case_id
stringlengths 8
8
| modality
stringclasses 6
values | indication
stringlengths 5
20
| mechanism
stringlengths 13
31
| target
stringlengths 3
13
| model_system
stringclasses 10
values | study_design
stringlengths 17
45
| efficacy_summary
stringlengths 28
65
| safety_summary
stringlengths 30
50
| pk_pd_summary
stringlengths 17
63
| biomarker_summary
stringlengths 11
47
| confounders
stringlengths 25
56
| data_gaps
stringlengths 33
59
| decision_options
stringclasses 1
value | expected_best_decision
stringclasses 3
values | expected_rationale_bullets
stringlengths 71
139
| red_flags_expected
stringlengths 31
71
| risk_level
stringclasses 3
values |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
PDC-0001
|
Small molecule
|
NAFLD
|
PPAR modulation
|
PPAR-delta
|
Mouse DIO
|
8w; n=12/arm; 3 doses; vehicle control
|
Liver fat -18% at top dose; no dose-response; ALT unchanged
|
No deaths; mild weight loss at top dose; AST +22%
|
Exposure high variance; Tmax delayed; trough below EC50 in 2/12
|
FGF21 +5% (ns); inflammatory panel mixed
|
Food intake not controlled; weight loss may drive effect
|
No histology scoring; no blinded readout
|
GO|HOLD|KILL
|
HOLD
|
Effect lacks dose-response|Possible confound from weight loss|AST rise needs clarification|PK variability undermines interpretation
|
No dose-response|Confounded efficacy|Liver enzyme signal|PK variability
|
Medium
|
PDC-0002
|
mAb
|
Oncology solid tumor
|
Receptor blockade
|
EGFR
|
PDX model
|
21d; n=8; weekly dosing; comparator anti-EGFR
|
Tumor growth inhibition 48% vs control; consistent across animals
|
No clinical signs; cytokines stable
|
Exposure consistent; target occupancy 85%
|
pERK down 60% post-dose; rebounds by day 7
|
PD sampling timing may bias rebound
|
No immunogenicity assessment; no combo data
|
GO|HOLD|KILL
|
GO
|
Efficacy consistent|Clean safety|Strong target engagement|PD matches MoA|PK stable
|
PD rebound risk|Missing immunogenicity
|
Low
|
PDC-0003
|
Small molecule
|
Alzheimer's
|
Kinase inhibition
|
GSK3beta
|
Rat
|
4w; n=10; oral daily; two doses
|
Cognitive score improved 9% (p=0.08); no clear dose separation
|
Sedation reported in 4/10 high dose; falls in 2/10
|
Brain exposure adequate; peripheral exposure high
|
pTau down 12% (ns); neuroinflammation unchanged
|
Sedation may inflate cognitive test artifacts
|
No chronic tox; no female animals
|
GO|HOLD|KILL
|
KILL
|
Efficacy weak and not significant|Dose separation absent|CNS side effects prominent|Biomarker shift minimal|Risk of symptomatic masking
|
Weak efficacy|CNS adverse effects|No dose separation
|
High
|
PDC-0004
|
siRNA
|
Hypercholesterolemia
|
RNA silencing
|
PCSK9
|
Non-human primate
|
Single dose; n=4; follow 8w
|
LDL -52% sustained to week 6
|
Platelets stable; mild injection reactions
|
Exposure consistent; durable effect
|
PCSK9 protein down 80% at week 2
|
Small n limits confidence
|
No repeat-dose tox; no off-target transcriptome
|
GO|HOLD|KILL
|
GO
|
Large durable LDL reduction|Safety acceptable|Mechanism confirmed|Clear translational precedent
|
Small n|Missing repeat-dose tox
|
Low
|
PDC-0005
|
Small molecule
|
Asthma
|
Inflammation modulation
|
CRTH2
|
Guinea pig
|
2w; allergen challenge; n=9
|
Airway resistance improved 25% at mid dose only; high dose worse
|
No deaths; mild diarrhea at high dose
|
Exposure increases with dose; high dose peak high
|
Eosinophils down 10% (ns)
|
High dose paradox suggests off-target
|
No receptor occupancy assay; no metabolite ID
|
GO|HOLD|KILL
|
HOLD
|
Non-monotonic efficacy suggests off-target|High dose worsening is concerning|Need occupancy and off-target profiling|GI tolerability issues
|
Non-monotonic response|Worsening at high dose|Missing occupancy
|
Medium
|
PDC-0006
|
Cell therapy
|
Oncology hematologic
|
Engineered T cells
|
BCMA
|
Mouse xenograft
|
n=10; single infusion; 30d follow
|
Complete response in 6/10; partial in 2/10
|
Cytokine spike day 2; 1 death likely CRS
|
Cells expand 50x then contract
|
IL6 high in responders and non-responders
|
Supportive care inconsistent
|
No dose-ranging; no mitigation strategy tested
|
GO|HOLD|KILL
|
HOLD
|
Strong efficacy but safety event|Need CRS mitigation plan|Dose-ranging required|Supportive care variability confounds safety read
|
Treatment-related death|CRS risk|No dose-ranging
|
High
|
PDC-0007
|
Small molecule
|
Type 2 diabetes
|
GLP1 agonism (oral)
|
GLP1R
|
Dog
|
4w; n=6; fed vs fasted arms
|
HbA1c trend -0.3; glucose AUC improved 12% only fasted
|
Nausea in 3/6; vomiting in 1/6
|
Huge food effect; exposure drops 70% fed
|
Insulin secretion markers mixed
|
Fed/fasted imbalance across groups
|
No formulation optimization; no exposure-response model
|
GO|HOLD|KILL
|
HOLD
|
Efficacy depends on fasting|Food effect undermines real-world use|Need formulation work|GI AEs present|Exposure-response unclear
|
Severe food effect|Real-world feasibility risk|GI AEs
|
Medium
|
PDC-0008
|
mAb
|
Autoimmune
|
Pathway inhibition
|
IL-23
|
NHP
|
Repeat dose q4w x3; n=5
|
Inflammation score down 40% sustained
|
No serious AEs; mild infections in 2/5
|
Exposure stable; ADA detected low titer
|
IL-17A down 55%
|
Mild infections may signal immunosuppression depth
|
No opportunistic infection panel; small n
|
GO|HOLD|KILL
|
GO
|
Clear efficacy|Manageable safety|PK stable|ADA low|Mechanism consistent
|
Infection signal|Limited infection workup
|
Low
|
PDC-0009
|
Small molecule
|
Oncology
|
DDR inhibition
|
ATR
|
Cell line + mouse
|
In vitro IC50 strong; in vivo 14d; n=10
|
Tumor reduction 15% only in one line; others none
|
Weight loss 12% in treated; neutropenia noted
|
Exposure exceeds IC90; narrow margin
|
pCHK1 down 70%
|
In vivo model may not match biomarker-defined population
|
No patient stratification hypothesis; limited combo testing
|
GO|HOLD|KILL
|
KILL
|
In vivo efficacy poor despite target engagement|Toxicity significant|Therapeutic window weak|No clear responder hypothesis
|
Poor efficacy|Toxicity|Narrow window
|
High
|
PDC-0010
|
Protein
|
Rare disease
|
Enzyme replacement
|
IDUA
|
NHP
|
Single dose; n=3; 6w follow
|
Substrate down 35% in plasma; tissue unknown
|
Hypersensitivity reaction in 1/3
|
Exposure adequate; half-life short
|
Biomarker response transient
|
Premedication not standardized
|
No tissue biodistribution; no repeat dosing
|
GO|HOLD|KILL
|
HOLD
|
Some biomarker effect but transient|Hypersensitivity risk needs plan|Need tissue readouts|Need repeat-dose data
|
Hypersensitivity|No tissue efficacy data
|
Medium
|
PDC-0011
|
Small molecule
|
Depression
|
NMDA modulation
|
NR2B
|
Rat
|
2w; n=12; behavioral assays
|
Behavioral improvement 20% across tests
|
No deaths; locomotor hyperactivity at high dose
|
Exposure OK; brain penetration high
|
BDNF up 30%
|
Hyperactivity may inflate behavioral outcomes
|
No chronic tox; abuse liability not explored
|
GO|HOLD|KILL
|
HOLD
|
Signal present but confounded by hyperactivity|Need assay controls|Need longer tox|Need liability assessment
|
Behavior confound|CNS activation risk
|
Medium
|
PDC-0012
|
Vaccine
|
Infectious disease
|
Neutralizing antibody induction
|
Spike antigen
|
Mouse
|
Prime-boost; n=15
|
Neutralization titers high; challenge protection 80%
|
Local reactogenicity moderate; no severe AEs
|
PK not applicable
|
Th1 skew; cytokines acceptable
|
Challenge model variability
|
No durability beyond 8w; no variant panel
|
GO|HOLD|KILL
|
GO
|
Strong protection|Safety acceptable|Immune profile supports protection|Clear next experiments identified
|
Durability unknown|Variant coverage unknown
|
Low
|
Clarus Preclinical Decision Coherence v0.1
What this dataset is
This dataset tests whether a model can make clear, disciplined preclinical decisions under realistic uncertainty.
It focuses on a single question.
Can the system decide GO, HOLD, or KILL
and justify that choice without inventing data or avoiding risk.
Why this matters in pharma
Preclinical failures are rarely due to missing data.
They fail because:
- Signals are weak but not named
- Confounders are ignored
- Momentum replaces judgment
This dataset measures whether a model can:
- Name uncertainty
- Surface red flags
- Resist optimistic drift
Dataset structure
Single CSV file.
- Downloads last month
- 21