--- license: cc-by-nc-4.0 language: - en tags: - synthetic - healthcare - gene-therapy - cell-therapy - car-t - crispr - aav - clinical-trials - immunogenicity pretty_name: "Gene Therapy Response Dataset: Multi-Modality Longitudinal Efficacy, Safety, Immunogenicity & Long-Term Follow-Up (Sample)" size_categories: - 1K **Not for clinical use.** This dataset is for ML development, benchmarking, schema prototyping, and education only. It must not be used to inform real patient care, trial design, or regulatory decisions. ## Unit of observation The unit is the **dosed subject**. The patient registry also retains AAV **screen failures** (nAb+) from the upstream eligibility gate, so the registry row count exceeds the dosed cohort. The five other tables key on `subject_id` for dosed subjects. ## Calibration anchors The engine is *benchmark-first*: each simulation parameter maps 1:1 to one validation metric, locked via exact index selection. Sample-level observed values (seed 42, ~500 dosed): | Metric | Observed | Target | Anchor | |---|---|---|---| | Modality coverage | 1.00 | =1.0 | all 5 modalities present | | Indication coverage | 1.00 | =1.0 | all 12 indications present | | Durable response rate | 0.620 | 0.57–0.67 | pooled pivotal response | | Year-5 durability rate | 0.710 | 0.65–0.77 | LTFU durability | | CRS rate (CAR-T, true) | 0.780 | 0.72–0.84 | Lee 2019 grading | | ICANS rate (CAR-T) | 0.323 | 0.26–0.38 | ASTCT 2019 grading | | SAE rate per patient-year | 0.373 | 0.32–0.50 | safety-model accounting | | AAV nAb exclusion rate | 0.300 | 0.27–0.33 | pre-screen eligibility gate | | CRISPR on-target edit separation | 31 pp | ≥15 (floor) | successful vs unsuccessful edit | | Transfusion-independence separation | ~0.88 | ≥0.40 (floor) | CRISPR hemoglobinopathy cohort | | B-cell aplasia (CAR-T specificity) | high | ≥0.30 (floor) | CD19-directed CAR-T | | AAV nAb titer separation | large | ≥3.0 (floor) | post-dose anti-capsid rise | | CRS-label leak (non-CAR-T) | 10 | 0–15 (disclosed bug) | see Limitations | | Hepatotox outside high-dose AAV | 0 | =0 (floor) | referential integrity | Validation: **Grade A+ (10.00/10)** across all six canonical seeds (42, 7, 123, 2024, 99, 1), deterministic. The engine additionally passes its own internal 12/12 benchmark suite on every seed. ## Tables Six relational CSVs keyed on `subject_id`: - **`hc_gen_007_patient_registry.csv`** — ~564 rows × 20 cols (dosed + nAb screen failures): indication (ICD-10), modality, product, demographics, baseline severity & scale, prior therapies, baseline nAb titer, HLA, conditioning, dose, infusion date, enrollment status. - **`hc_gen_007_efficacy_outcomes.csv`** — ~6,000 longitudinal visits × 15 cols: VCN, transgene expression, primary-endpoint trajectory (scale-specific), response category, durability, motor-milestone / transfusion-independence / ABR-reduction flags, BCVA change, MRD, on-target edit %. - **`hc_gen_007_adverse_events.csv`** — ~10,000 AEs × 18 cols: MedDRA PT/SOC, CTCAE grade, seriousness, relatedness, action, outcome, CRS (Lee grade), ICANS (ASTCT grade), hepatotoxicity, TMA, special-interest flags. - **`hc_gen_007_immunogenicity_lab.csv`** — ~6,000 labs × 16 cols: anti-capsid & anti-transgene antibody titers, ALT/AST/bilirubin/LDH, ferritin, IL-6, troponin, CD3/CD19, B-cell aplasia, complement C3. - **`hc_gen_007_long_term_followup.csv`** — ~3,700 annual records × 13 cols: durability, secondary malignancy, integration-site concern, VCN/transgene persistence, late AE, mortality, off-target events (to Year 15 for integrating vectors). - **`hc_gen_007_trial_summary.csv`** — ~36 aggregate rows × 14 cols: per (indication, modality, cohort phase) response rate, AE/SAE rates per patient-year, CRS/ICANS/mortality/dropout rates. ## Loading ```python import pandas as pd registry = pd.read_csv("hc_gen_007_patient_registry.csv") efficacy = pd.read_csv("hc_gen_007_efficacy_outcomes.csv") aes = pd.read_csv("hc_gen_007_adverse_events.csv") # CRS rate among CAR-T (exclude the disclosed hepatotox mislabel) cart = set(registry.loc[registry.modality=="CAR_T","subject_id"]) crs = set(aes.loc[(aes.is_crs=="Y") & (aes.is_hepatotoxicity!="Y"),"subject_id"]) & cart print(len(crs)/len(cart)) ``` ```python from datasets import load_dataset registry = load_dataset("xpertsystems/hc-gen-007-sample", "patient_registry") ``` ## Use cases - Multi-modality gene/cell-therapy response and durability modeling. - CAR-T toxicity (CRS/ICANS) prediction and grading-aware safety models. - AAV immunogenicity / nAb eligibility and hepatotoxicity-risk modeling. - CRISPR on-target editing and transfusion-independence outcome modeling. - Long-term-follow-up survival, integration-site, and durability analysis. - Relational/longitudinal ML pipeline prototyping over a six-table star schema. ## Limitations (honestly disclosed) - **Disclosed engine bug: hepatotoxicity events are mislabeled `is_crs="Y"`.** The high-dose-AAV hepatotoxicity special event is emitted with the CRS flag set true (the `is_crs_flag` tuple field is `True` for `Hepatocellular injury` rows; it should be `False`). This leaks ~10 non-CAR-T `is_crs="Y"` rows per 500 subjects. The engine's own Metric 4 hides this by intersecting with CAR-T IDs. **To compute a correct CAR-T CRS rate, exclude rows where `is_hepatotoxicity=="Y"`** (as in the loading example). The sample scorecard surfaces this as an explicit `crs_label_leak_noncart` metric rather than hiding it. Recommended full-product fix: set the CRS flag to `False` on the hepatotox tuple. - **Benchmark-first by construction.** Headline rates (durable response, CRS, ICANS, nAb exclusion, Year-5 durability) are locked near-exactly to their `BENCH` targets via index selection, so cross-seed variance is minimal — realistic for a calibrated simulator but not a substitute for empirical trial heterogeneity. - **SAE rate runs slightly below its target.** Observed ~0.37/PY vs `BENCH` 0.42; the special-event SAE accounting under-fills the quota. Within the engine's own ±0.08 tolerance. - **Synthetic, archetype-based.** Modality/indication mappings and product archetypes are illustrative, not specific approved products. - **Marginal calibration, not full joint fidelity.** Headline rates and the engineered structural separations (CRISPR edit gap, transfusion-independence concentration, CAR-T B-cell aplasia, AAV nAb rise) are anchored; higher-order correlations beyond those engineered are not independently validated. - **Small-sample note.** At ~500 dosed subjects, cohort-specific rates (CRS/ICANS within CAR-T, hepatotox within high-dose AAV) are computed on modest subgroups; scorecard ranges accommodate this and structural floors are weighted to dominate. ## Commercial / full version | | Sample (this) | Full (commercial) | |---|---|---| | Dosed subjects | ~500 | 5,000 (configurable) | | Tables | 6 (full schema) | 6 (full schema) | | Formats | CSV | CSV / Parquet | | Engine fixes | as-is, bug disclosed | `is_crs` hepatotox mislabel patched on request | | Seeds / reproducibility | 6 canonical | Unlimited | | License | CC-BY-NC-4.0 | Commercial | | Support | — | SLA, custom modalities/indications, vertical extensions | Contact **pradeep@xpertsystems.ai** · https://xpertsystems.ai ## Citation ```bibtex @dataset{xpertsystems_hcgen007_2026, title = {HC-GEN-007: Synthetic Gene Therapy Response Dataset --- Multi-Modality Longitudinal Efficacy, Safety, Immunogenicity & Long-Term Follow-Up (Sample)}, author = {XpertSystems.ai}, year = {2026}, publisher = {Hugging Face}, note = {Synthetic data. Not for clinical use. Fictional products. Calibration anchors: Lee et al. 2019 (CRS grading); ASTCT 2019 (ICANS grading); FDA 2020 Long-Term Follow-Up Guidance; AAV/CRISPR/CAR-T/lentiviral pivotal-trial response and durability benchmarks. Disclosed defect: hepatotoxicity special events mislabeled is_crs=Y.}, url = {https://huggingface.co/datasets/xpertsystems/hc-gen-007-sample} } ```