All Datasets
Everything built so far, in one list. Narrow it down by domain, or sort it by whatever you are actually judging on.
bTBI — Blast Traumatic Brain Injury in Blast-Exposed Service Members
20,000 synthetic blast-exposed service members: acute injury severity (GCS, LOC/AOC/PTA, CT imaging — Marshall/Rotterdam), serial MACE-2/ImPACT/VOMS cognitive testing, and 6-12 month PTSD, sleep, headache and functional outcomes (540,000 longitudinal rows), mapped to OMOP CDM v5.4. 105/133 parameters VERIFIED against 75 sources; 112/112 validation checks passed; baseline recovery model AUC 0.72 sits inside the published real-data range. Free sample included.
Hospital Procure-to-Pay Recovery Ledger
A relational test ledger linking hospital purchase orders, receipts, invoices, approvals, bank payments, returns, credit notes and rebates: 5,000 invoices across 33 tables with a 470-finding, 450-case answer key spanning five designed recovery-fault families. Built for recovery-audit, three-way-match and reconciliation software testing, not for clinical or patient-level use.
Skin Cancer Clinical Features & Staging
10,000 synthetic skin lesions on 8,221 adults, each read independently by three observer types (GP, dermatologist, dermatologist with dermoscopy), with the true diagnosis, Breslow thickness, ulceration and AJCC 8 T-category stored next to every reading. 4 Parquet tables, 37,497 observations, 35 audited parameters, 46 sources, bit-for-bit reproducible.
Antibiotic Resistance Feedback 100k
100,000 virtual adults on piperacillin-tazobactam where the drug corrupts the creatinine it is dosed from — closed-loop dosing, resistance selection and AKI, two calibrated scenarios, observed records and ground truth side by side. 18 Parquet tables, 6.89M rows, 43 sources, 202 audited constants, bit-for-bit reproducible.
HCTO — Hemorrhage Control and Tourniquet Outcomes
25,000 synthetic combat casualties with extremity and junctional hemorrhage — 81 fields plus 323,704 five-minute vital-sign rows covering tourniquet application and effectiveness, hemostatic packing, TXA, Role 2 labs, transfusion by echelon and 24-hour / 30-day outcomes. 129 parameters anchored to 78 sources, 3 ML tasks benchmarked against clinical baselines, OMOP CDM v5.4 mapping.
FTW — Ferroptosis Timing Window (EGFR+ NSCLC)
In-silico EGFR-TKI-resistant NSCLC dataset from a five-compartment ODE model: a 5,000-patient 15-axis screen, a 3,000-patient virtual cohort and a 7-dose scan, with paired TKI vs TKI + ferroptosis-agent arms — 7 Parquet files, 43,560 rows, every parameter literature-audited.
Combat PTSD & Veteran Mental Health Cohort
30,000 synthetic U.S. veterans followed for 24 months (150,000 assessment rows) with combat-exposure-driven PTSD, depression, anxiety, substance-use, suicidality, treatment-response and disability outcomes — PCL-5 / PHQ-9 / GAD-7 / C-SSRS instruments, ML-ready labels and a full OMOP CDM v5.4 mapping. 117 of 118 parameters VERIFIED against 44 published sources; 11/11 validation checks and 10/10 preregistered predictions passed; free 500-veteran sample included.
Ukraine War Health Atlas
Harmonized oblast×month panel linking attacks on healthcare, facility status, primary-care utilization, displacement, and refugee flows across 25 Ukrainian oblasts over 67 months of war.
Vancomycin Renal Feedback 100k
100,000 virtual adults on vancomycin with renal function as a state variable — closed-loop nephrotoxicity, two calibrated scenarios, observed records and ground truth side by side. 18 Parquet tables, 9.27M rows, 57 sources, 242 audited constants, bit-for-bit reproducible.
Coherent Patient Records 1M (with Generator & Validation Scripts)
1,000,000 synthetic U.S. inpatient EMR records in 43 columns where diagnoses drive vitals, labs, medications, ICU use, length of stay, mortality and costs, plus three hidden ground-truth columns. Generator script and a 50-check validation script included; 79 sourced parameters. Free 10,000-row sample.
Healthcare Cost & Claim Structure (200,000 Episodes)
200,000 synthetic medical encounters — inpatient, outpatient and emergency — with about 1.18 million itemized claim lines, payer and patient share, denials and appeals, and a ground-truth internal cost per episode (true_episode_cost). 18 coherence rules, no CPT licensing issues, free sample included.
Stroke Time-to-Treatment Counterfactual (200,000 Patients) + Generator
200,000 synthetic stroke patients with counterfactual treatment-timing trees: every treated patient carries tPA and/or thrombectomy branches delivered at the actual time and 15, 30 and 60 minutes earlier — 302,400 branch rows, 7 latent ground-truth columns, 90-day mRS outcome. Every parameter literature-sourced, generator included, 40/40 primary anchors passed; free sample included.
MATERNA-ATLAS — Harmonized Multimodal Maternal-Fetal Health Atlas
12,860 linked maternal-fetal episodes from 9 open-access sources — tabular risk records, 552 raw CTG and 326 raw EHG waveforms, plus two fetal-ultrasound benchmarks (HC18, FETAL_PLANES). Group-safe splits, leave-one-source-out folds, reference baselines for every arm, and every quality issue flagged rather than hidden. Free sample included.
GLP-1 Virtual Cohort v1
20,000 synthetic patients on 6 GLP-1/GIP/amylin/glucagon molecules (semaglutide, tirzepatide, orforglipron, retatrutide, eloralintide, CagriSema), followed weekly for 104 weeks across 6 scenario branches on the same patients (as-treated + switch/dose-change/holiday/early-stop/never-treated counterfactuals) — 12.48M weekly records, two-compartment PK, 113/113 validation gates passed, free stratified sample included.
Cardiovascular Risk & Heart Failure Cohort (10,000 Patients, 10-Year Follow-Up)
10,000 synthetic patients followed for up to 10 years across three cardiovascular strata (primary prevention, secondary prevention, heart failure) — 77,164 annual visit records, SCORE2 / ASCVD Pooled Cohort Equations-anchored baseline risk, and MAGGIC-calibrated heart-failure outcomes. Built for cardiovascular risk-prediction modeling, HF research, and synthetic-control / methods-development work. 69/69 validation gates passed; free stratified sample included.
RegTrace v1 — FDA Regulatory Trajectory Dataset for Reinforcement Learning
Real FDA drug-application regulatory trajectories — submissions, Complete Response Letters, and final outcomes — packaged as verifiable-reward RL episodes and SFT traces for training regulatory-decision AI agents.
Trauma Hemorrhage & Coagulopathy Cohort
10,000 synthetic severe-hemorrhage trauma patients simulated under three counterfactual resuscitation strategies with a literature-calibrated, 8-state mechanistic ODE model — 21.6M profile rows built for resuscitation-policy and coagulopathy-prediction algorithm development.
Acid Trap Dataset (Virtual Screening, Lung Cancer)
A fully synthetic, ODE-modeled dataset simulating four-valve intracellular pH-collapse ("Acid Trap") therapy in lung cancer — 179,351 rows spanning a mechanistic virtual screen, TCGA-anchored virtual patients, and a 10,000-profile synthetic cohort.
Immunotherapy Clonal Dynamics Cohort
10,000 fully synthetic cancer patients simulated under three counterfactual immunotherapy regimens on identical underlying biology, with mechanistic tumor-immune-drug dynamics and literature-anchored parameters.
Sepsis Resuscitation Cohort + Generator
10,000 fully synthetic septic patients simulated under three counterfactual resuscitation strategies from one shared physiology, delivered together with the full mechanistic generator that produced them — 21.6 million hemodynamic profile rows, bit-for-bit reproducible.
Tumor Growth & Chemotherapy Response Cohort (10,000 Patients, 3 Counterfactual Arms)
Synthetic oncology cohort of 10,000 patients across four cancer types, each simulated on three counterfactual chemotherapy schedules — MTD, dose-dense, and metronomic — plus an oracle label for which arm would have worked best. No real patient data anywhere in the generation.
MAUDE Atlas - FDA Medical Device Safety Database (25.7M Reports)
MAUDE-Atlas v1.0.0 is a global medical-device safety atlas built from the FDA MAUDE database. 25.7M real post-market reports, enriched with GUDID, recall, and classification data, coded with official FDA MDR terminology. Designed for post-market surveillance, signal detection, and ML benchmarking. Includes deduplicated clusters, leakage-controlled splits, and a free 100k-report sample.
Warfarin Pharmacogenomics Dosing Cohort (Data + Full Generator Code)
Synthetic benchmark cohort: 10,000 patients × 3 counterfactual warfarin dosing strategies, full longitudinal INR/PK/PD profiles (0–120 d, step 0.5 d), stochastic bleeding/thrombosis events, oracle optimal dose. Includes full Python generator code, spec, audit trail. Free 500-patient sample.
Vancomycin Precision-Dosing Cohort (10,000 Patients)
Pharmacokinetic dataset of 10,000 adult patients for Bayesian dosing algorithm development, TDM strategy evaluation, and PK/PD model benchmarking.
FK2D Tumor Growth Phantom Dataset
Synthetic collection of 1,000 simulated 2D glioma growth cases (500 LGG / 500 HGG) generated via Fisher–Kolmogorov reaction-diffusion PDE. Contains 8,000 total timepoints with float32 cell density fields, 3-channel segmentation masks, growth trajectories, and T2-like grayscale PNGs on a 256×256 mm grid.
Huntington's Disease Cohort Dataset
A longitudinal synthetic cohort of 5,000,000 patients, structured in strict adherence to the OMOP Common Data Model (CDM) v5.4. Features an epidemiologically enriched subset of 400 true-positive Huntington's Disease profiles with genetic CAG repeat validation, family pedigree graph linkages, and pharmacotherapy exposure spanning 2016–2026.
Immunosuppression PK: Tacrolimus & Cyclosporine
A fully synthetic corpus of 25,000 solid-organ transplant recipients (825,000 TDM visits) modeling tacrolimus and cyclosporine exposure. Provides exact latent ground truth (true clearance, unbound exposure) paired with noisy clinical observations and calibrated rejection risk.
Combat Triage and Trauma Outcomes (CTTO)
A 30,000-record synthetic combat casualty registry modeled on US DoD JTTR schemas, featuring 78 variables across care timelines, TCCC interventions, lab data, TRISS mortality, and full OMOP CDM v5.4 mapping.
MitoKinetics-20M — Synthetic Mitochondrial & mTOR Pathway Dataset
A synthetic longitudinal dataset modeling mitochondrial therapeutic response to mTOR-pathway modulation across a 10-year treatment window in a cohort of 1,000,000 virtual patients (20,000,000 rows, Apache Parquet).
Early Deterioration Calibrated Constants Pack
A fully documented, primary-source-verified parameter table containing 104 physiological constants for simulating early clinical deterioration and the NEWS2 scoring algorithm.
Early Deterioration Dataset
A physiologically-grounded synthetic ward population of 600 patients (11,289 observations) for early warning score research, deterioration prediction, and clinical AI training based on coupled ODE models and the NEWS2 algorithm.
Sepsis-AKI Semi-Competing Risk Cohort
A fully synthetic critical-care cohort (500 patients) modelling the interaction between antibiotic-induced acute kidney injury (AKI) and death from sepsis using a semi-competing risks framework.
Gentamicin Pharmacokinetics Dataset
A physiologically grounded synthetic dataset (600 patients) for intravenous gentamicin therapeutic drug monitoring (TDM). Features ground-truth CL/Vd for parameter-estimation training.
Synthetic CBC Abnormality Dataset
50,000 fully synthetic Complete Blood Count (CBC) records calibrated to published population reference data. Includes 14 columns and 4 diagnostic labels (normal, anemia, infection, leukemia flag) delivered in Apache Parquet format.
MH Trigger Response Constants Pack
Standalone, audited constant set for teams building their own RyR1-cascade simulation rather than consuming the full scenario dataset.
Multicompartmental Pharmacokinetics with Renal and Hepatic Impairment (60k patients)
A reference corpus of 60,000 synthetic patients receiving an intravenous drug against a time-varying background of renal and hepatic function, containing both true individual PK parameters and a realistic observed TDM layer.
Malignant Hyperthermia & Perioperative Crisis
Mechanistic synthetic dataset of 27 perioperative malignant hyperthermia scenarios (235,467 telemetry rows) modeling RyR1 calcium-release cascade coupled with human hemodynamics and dantrolene PK/PD intervention.
TriageFlow-11: Synthetic Clinical Simulation
A synthetic emergency department queue simulation with 545,231 patients across 11 countries. Includes physiological vitals, resource contention, and four distinct layers: counterfactual twins, known-magnitude bias, rosetta cross-scale labels, and a crisis dial.
Rare Disease Natural History Engine
Mechanistically calibrated synthetic registry data across three rare diseases (DMD, Huntington's, Gaucher Type 1)
Trauma: Hemorrhagic Shock + Tourniquet Time
Synthetic mechanistic dataset containing time-series vital-sign telemetry and outcome labels for limb ischemia-reperfusion under hemorrhagic shock. Includes 28 synthetic patients and 13,468 rows. Strictly for AI training and methodological demonstration, not for clinical decision-making.
GliomaFlow-3D: Synthetic Longitudinal Brain Tumor MRI Dataset
Fully synthetic, longitudinal MRI dataset simulating brain tumor (glioma) growth in 2,000 virtual patients across 3 time points (6,000 total MRI scans) with exact PDE ground truth and Bloch-physics signals.
ICU Sepsis Telemetry Dataset
Physiologically-grounded synthetic ICU telemetry dataset for sepsis (1,000 patients, 120-hour window). Generated via differential-equation models with zero privacy risk.
SIRS/MODS Constants Package for Sepsis Modeling
A source-cited, individually verified parameter set of 57 constants with literature citations and explicit calibration statuses for building physiological sepsis, SIRS, and MODS models.