CureGraph — Biomedical Evidence, Knowledge Graph, Hypothesis & Experiment Workspace
CureGraph organizes peer-reviewed scientific evidence from PubMed/NCBI and ClinicalTrials.gov, maps provenance-linked biomedical knowledge graphs, computes transparent and overridable evidence-confidence scores, and gates wet-lab and computational experiments behind rigorous ethics checks.
Mandatory Research Integrity Notice
“For research use by qualified professionals. Not for clinical decision-making or medical advice.”
CureGraph never provides patient-specific medical advice, never recommends drugs or dosages, never claims that association proves causation, never fabricates citations, and never promises that AI can cure a disease.
Seeded Demo Workspace
Pancreatic Cancer (PDAC)
KRAS G12D, TP53, CDKN2A, SMAD4, MAPK/ERK
Verified Citations
7 PubMed + 3 NCT Trials
Zero hallucinated citations; real PMID & NCT links
Hypothesis Score
78 / 100 Weighted
Transparent 6-factor formula · PI overridable
Ethics & Governance
8-Check Gate
5 mandatory checks required before Running stage
Scientific Workflow Architecture
Discover → Screen → Synthesize → Hypothesize → Validate
Every stage preserves provenance, distinguishes supporting from methodological-concern records, and enforces human scientific oversight.
Discover
Live PubMed / NCBI E-utilities & ClinicalTrials.gov APIv2 search with labeled demo fallbacks.
Query peer-reviewed literature and clinical study registries without ever inventing or fabricating citations.
Screen
Classify papers by role: Supporting, Contradicting, Methodological Concern, or Contextual.
Separate primary experimental cohorts from clinical background reviews and flag confounders such as low neoplastic cellularity.
Synthesize
Map biomedical entities (Disease, Gene, Protein, Pathway, Compound, Biomarker, Trial, Experiment).
Every node and edge carries explicit PMID or NCT provenance and warns that association is not proof of causal dependence.
Hypothesize
Compute transparent, overridable Evidence-Confidence Scores with explicit falsification criteria.
Weighted summary of source quality, replication, relevance, biological plausibility, recency, and contradiction burden — never a truth probability.
Validate
Prioritize wet-lab & computational experiments behind an 8-point Ethics & Governance Gate.
Blocks transition to the Running stage until human scientific oversight, privacy minimization, consent, reproducibility, and misuse prevention are verified.
Seeded Workspace Showcase
Pancreatic Cancer Target Discovery — Demo
Explore the pre-populated PDAC workspace: KRAS G12D / TP53 / CDKN2A / SMAD4 / MAPK knowledge graph, annotated PubMed literature, ClinicalTrials.gov comparator, and ethics-gated experiments.
PDAC Core Driver & Pathway Topology
Click any node in the workspace to inspect PMID / NCT provenance. Associations do not prove uniform causal dependence.
Pancreatic ductal adenocarcinoma. Demonstration disease node — not a diagnosis.
Provenance: Demo graph, PDAC literature
Oncogene mutated in ~90% of PDAC, most often G12D.
Provenance: Jones 2008 PMID 18772397; TCGA 2017 PMID 28803728
Tumor suppressor frequently inactivated in PDAC.
Provenance: Jones 2008 PMID 18772397
Encodes p16INK4a; frequent loss in PDAC, deregulating G1/S.
Provenance: Jones 2008 PMID 18772397
TGF-β pathway tumor suppressor frequently lost in PDAC.
Provenance: Jones 2008 PMID 18772397
Constitutively active KRAS oncoprotein; candidate direct target.
Provenance: Waters & Der 2018 PMID 29229669
CDKN2A product; CDK4/6 inhibitor at G1/S.
Provenance: Demo graph
Effector pathway downstream of mutant KRAS. Pathway association is not proof of uniform dependence.
Provenance: Waters & Der 2018 PMID 29229669
KRAS G12D dependence of MAPK output in PDAC models
Research Question: Does selective KRAS G12D inhibition reduce MAPK signaling and viability in KRAS G12D-mutant PDAC models relative to KRAS wild-type controls?
Falsification Criteria: If phospho-ERK and viability are unchanged in validated G12D-mutant models at exposures that engage KRAS G12D, the hypothesized MAPK dependence is falsified for those models.
overall = 100 × clamp((0.22·sourceQuality + 0.18·replication + 0.18·relevance + 0.16·biologicalPlausibility + 0.12·recency) / 0.86 − 0.14·contradictionBurden, 0, 1)
PubMed / NCBI Evidence Cards
Real PMIDs · Never FabricatedCore signaling pathways in human pancreatic cancers revealed by global genomic analyses
Note: Establishes KRAS, TP53, CDKN2A, SMAD4 as core PDAC pathways. Association in tumors is not proof of causal dependence on any single target.
Whole genomes redefine the mutational landscape of pancreatic cancer
Note: Replicates core drivers and adds DNA-repair subsets that may modify therapeutic hypotheses.
Genomic analyses identify molecular subtypes of pancreatic cancer
Note: Subtype context (squamous vs progenitor) may stratify MAPK dependence hypotheses.
Integrated Genomic Characterization of Pancreatic Ductal Adenocarcinoma
Note: Low neoplastic cellularity can confound genomic calls. Interpret KRAS wild-type calls cautiously.
ClinicalTrials.gov 3-Study Comparator
Registry ≠ Proof of Safety/EfficacyStudy of MRTX1133 in Patients With Advanced Solid Tumors Harboring a KRAS G12D Mutation
Sponsor: Mirati Therapeutics Inc. · Interventions: MRTX1133 · Outcomes: Dose-limiting toxicities, Objective response rate
Olaparib in gBRCA Mutated Pancreatic Cancer Whose Disease Has Not Progressed on First Line Platinum-Based Chemotherapy (POLO)
Sponsor: AstraZeneca · Interventions: Olaparib, Placebo · Outcomes: Progression-free survival
S1505: Combination Chemotherapy or Gemcitabine Hydrochloride and Paclitaxel Albumin-Stabilized Nanoparticle Formulation Before Surgery in Treating Patients With Pancreatic Cancer
Sponsor: SWOG · Interventions: mFOLFIRINOX, Gemcitabine, Nab-paclitaxel · Outcomes: Overall survival, Resection rate
Ethics-Gated Experiment Board
Experiments cannot enter Running until all 5 required ethics checks are verified.
In vitro KRAS G12D inhibition panel
A. Chen, PI · PDAC cell lines (G12D vs WT)
Endpoint: pERK (6–24 h), CellTiter-Glo (72–120 h)
No human subjects. Cell-line work under institutional biosafety.
Organoid phospho-proteomics after KRAS G12D blockade
J. Okonkwo, Computational · Patient-derived PDAC organoids
Endpoint: pERK, phospho-proteome, ctDNA-analog mutant allele fraction in supernatant
De-identified organoids; confirm IRB coverage before Running.
Combination screen: KRAS G12D + CDK4/6 inhibition
M. Reyes, Translational · CDKN2A-loss vs intact organoids
Endpoint: Synergy (Bliss), apoptosis, senescence markers
Ethics checklist not started — Running stage blocked.
PDX MAPK pathway engagement study
A. Chen, PI · KRAS G12D PDX
Endpoint: Tumor pERK, volume (exploratory), body-weight safety
IACUC not yet filed — Running is blocked until remaining ethics checks complete.
Lab & Principal Investigator Onboarding
Request CureGraph Researcher Access
Request a dedicated CureGraph tenant for your academic lab, translational oncology group, or biotech R&D team. All workspaces enforce strict provenance auditing, local no-PHI CSV/JSON screening, and ethics-gated experiment progression.
- • Data Sources: PubMed / NCBI E-utilities, ClinicalTrials.gov APIv2, NCI GDC metadata, local browser-only CSV/JSON (no PHI).
- • Roles Supported: Principal Investigator (PI), Computational Biologist, Translational Scientist, Ethics & Governance Reviewer.
- • Self-Service Setup: Already have a GitHub organization? Start your lab workspace at /start.