Immunotherapy Clonal Dynamics Cohort

This Zenodo record contains a free 100-patient sample only — not the full 10,000-patient dataset. Full dataset (10,000 patients, full technical documentation): https://sentineldata.com.ua/dataset/immunotherapy-clonal-dynamics-cohort Sample composition (true subset of the full cohort, not a separate generation): patients_sample.parquet — 100 rows; treatment_regimens_sample.parquet — 300 rows; outcomes_sample.parquet — 300 rows; clonal_profiles_sample.parquet — 72,300 rows. Dataset Architecture & Specifications The Sentinel Immunotherapy Clonal Dynamics Cohort is a fully synthetic, mechanistically generated benchmark of tumor clonal dynamics under immune checkpoint inhibition (nivolumab, pembrolizumab, atezolizumab). Each of 10,000 synthetic patients is simulated under three counterfactual treatment arms on identical underlying biology, giving paired ground-truth comparisons no real-world trial can produce. 47.2% of generation constants are literature-verified (VERIFIED tier), the remainder are documented literature-anchored estimates (ESTIMATE_SOURCED), with zero unsourced values. Synthetic Patients 10,000 (3 counterfactual arms each) Treatment Regimens 30,000 (continuous / adaptive stop-resume / early aggressive) Clonal Profile Rows 7,230,000 (0-120 days, 0.5-day step) Data Schema Columnar Parquet (Snappy), 100% reproducible via Master Seed 20260828 Tumor-Immune-PK Model & Governing Equations Each patient is a deterministic 5-state ODE system (sensitive clone, resistant clone, immune effector population, two-compartment drug PK) integrated with fixed-step RK4 on a 0.5-day grid. Drug clearance is time-dependent, reflecting checkpoint-inhibitor target-mediated drug disposition. dV_sens/dt = λs·V_sens·(1 − V_total/K) − α·I_eff·V_sens − κ·C_c·V_sens dV_res/dt = λr·V_res·(1 − V_total/K) − αr·I_eff·V_res dI_eff/dt = σ − δ·I_eff + (ρ0 + β·C_c)·I_eff·V_total/(V_total + K_half) dC_c/dt = −(CL(t)/V1)·C_c − (Q/V1)·(C_c − C_p) dC_p/dt = (Q/V2)·(C_c − C_p) CL(t) = CL·(1 − emax·t^Hill / (T50^Hill + t^Hill)) Stratification & Phenotypic Complexity High Immune (19.5%): elevated effector influx and predation rate. Low Immune (20.4%): reduced effector influx and predation rate. High Resistance (19.8%): elevated baseline resistant-clone fraction. Low Resistance (20.2%): minimal baseline resistant-clone fraction. Fast Progressor (20.1%): accelerated sensitive-clone growth rate. Validation Metrics Parameter Provenance: 34/72 constants directly verified from published sources (47.2%), 38/72 sourced estimates with documented derivation, 0% unsourced. ORR Calibration: continuous-arm response rates match label-anchored targets within ±0.03 (melanoma 0.376 vs 0.365, NSCLC 0.169 vs 0.172, renal 0.214 vs 0.215). Reproducibility: 31/31 cohort-level checks and 12/12 model-level unit tests pass; full cohort reproduces bit-identically from the shipped code and seed. No Clinical Claims Synthetic data. Not for clinical use, dosing decisions, or claims of clinical efficacy. Parameters are literature-anchored but the system is a deliberately simplified mechanistic benchmark. License: CC BY 4.0. Version: 1.0.