xpertsystems/hcneu010-sample

HC-NEU-010 — Huntington's Disease (HD) Dataset (Sample)

A schema-identical preview of HC-NEU-010, the XpertSystems.ai synthetic longitudinal Huntington's Disease cohort dataset for clinical trial research, HTT-lowering ASO trial design, mHTT biomarker validation, CAG-driven progression modeling, and HD-specific machine learning. The full product covers 5,000 patients × 16 semi-annual visits. This sample is HF-sized at 200 patients × 16 semi-annual visits.

Built by XpertSystems.ai — Synthetic Data Platform Contact pradeep@xpertsystems.ai · xpertsystems.ai License CC-BY-NC-4.0 (sample); commercial license available for the full product.

---

What HC-NEU-010 does — and how it completes the genetic-neurodegeneration coverage

HC-NEU-010 is the tenth Healthcare / Neurology SKU in the XpertSystems catalog. Huntington's Disease is uniquely commercially positioned for synthetic data work:

• Fully-penetrant autosomal dominant — CAG ≥36 in HTT gene causes 100% disease (eventually), making HD the canonical disease for gene therapy and ASO drug development
• Tractable but rare — only ~30,000 symptomatic US patients + ~200,000 at-risk individuals, so real-world clinical trial cohorts are small (ENROLL-HD globally ~30,000), creating strong demand for synthetic data augmentation
• Three failed major drug trials in the last 5 years (Tominersen by Roche/Ionis, Branaplam by Novartis, PRX-12 by Prilenia) — pharma R&D needs better predictive modeling, and synthetic data is genuinely useful for trial-design simulation
• Tominersen is being retried in lower-dose / cognitively-defined subgroups (GENERATION-HD2) — making HD biomarker + trajectory data acutely valuable right now

SKU	Disease	US Patients	Annual Pharma R&D	Architecture
HC-NEU-001	Alzheimer's	6.9M	$8B	Single longitudinal
HC-NEU-002	Parkinson's	1.0M	$5B	Single longitudinal
HC-NEU-003	Epilepsy	3.4M	$3B	Cross-sectional
HC-NEU-004	Multiple Sclerosis	1.0M	$6B	Multi-table relational
HC-NEU-005	Stroke	7.0M	$3B	Cross-sectional
HC-NEU-006	Migraine	39M	$5B+	Cross-sectional
HC-NEU-007	ALS	30K	$2-3B	Single longitudinal
HC-NEU-008	TBI	3.5M	$2B	Cross-sectional
HC-NEU-009	Dementia (10-subtype)	7.0M+	$10B+	Multi-subtype longitudinal
HC-NEU-010	Huntington's	30K	$1-2B	Single longitudinal

HD + ALS + MS together form the "rare-but-high-investment monogenic/ genetic neurodegeneration cluster" — each has small patient populations but enormous per-patient pharma R&D spend due to gene therapy / ASO / cell therapy investment intensity.

This is the substrate HD pharma R&D teams (Roche, Novartis, uniQure, Wave, PTC Therapeutics, Prilenia), HD natural history registries (ENROLL-HD, REGISTRY), mHTT biomarker labs, CAG-prognostic modeling researchers, and HD-specific machine learning teams have been waiting for: a coherent longitudinal HD dataset where CAG repeat × disease stage × UHDRS motor × cognitive × imaging × NfL × mHTT all interact with ENROLL-HD / TRACK-HD / PREDICT-HD / Byrne 2018 NfL- grade calibration.

Buyer Persona	Use Case
HD Pharma R&D	HTT-lowering ASO comparator, dose-finding simulation
Gene Therapy Programs (uniQure, Wave)	Target-engagement (mHTT) modeling
ENROLL-HD / REGISTRY Analytics	Comparable cohort outcome research
CHDI Foundation	HD natural history augmentation
Tominersen Re-Trial Programs	Cognitively-defined subgroup enrichment
HD Biomarker Validation	NfL + mHTT longitudinal trajectory ML
Pre-Manifest HD Research	PREDICT-HD comparable cohort
HD Cognitive Reserve Research	Apathy + irritability + depression ML
HD Family Counseling Programs	Risk stratification by CAG count
Caudate Atrophy AI Imaging	TRACK-HD comparable atrophy ML

---

What's inside

Single wide longitudinal dataframe, multiple semi-annual visits per patient over 8-year follow-up.

Output	Rows (sample)	Columns	Size
HC_NEU_010_dataset.csv	~2,120	96	~1.1 MB

Row count varies (~2,122 at seed 42) due to mortality dropout and study dropout across the 8-year follow-up.

Schema provided in HC_NEU_010_schema.json.

Module structure (96 columns)

Module	Cols	Coverage
Visit metadata & treatment	10	patient_id, site, visit_number, visit_date, years_from_baseline, disease_stage, age, sex, education, treatment_arm, adherence
Genetics	11	CAG allele 1 (pathogenic) + allele 2, CAP score, somatic instability, predicted age of onset, MSH3 + PMS2 variants, family history, de novo flag, testing method
UHDRS Motor	12	TMS total, chorea, dystonia, bradykinesia, gait, tandem walking, finger tapping, pronate/supinate, rigidity, dysarthria, dysphagia, diagnostic confidence
Cognitive	11	SDMT, Stroop word/color/interference, Trail Making A/B, verbal fluency (letter + category), MoCA, MMSE, composite, annual decline
Psychiatric (UHDRS-Behavioral)	10	UHDRS-B total, PHQ-9, GAD-7, apathy, irritability, OCD, psychosis, suicidality, sleep, psychiatric hospitalization
Functional & QoL	9	TFC, TFC stage, Independence Scale, FAS, employment, living situation, nursing home, caregiver burden, HD-QoL
Clinical	6	BMI, weight change, dysphagia severity, falls, death flag, cause of death
Imaging	11	caudate, putamen, striatum, pallidum, whole brain, cortical thickness (frontal + temporal), corpus callosum FA, CST FA, caudate atrophy %, MRI metadata
Biomarkers	11	plasma + CSF NfL, plasma + CSF mHTT, mHTT detected, tau, GFAP, YKL-40, IL-6, BDNF, NfL annual change

---

Calibration sources

Every distribution is anchored to named clinical references. The headline anchors are ENROLL-HD (CHDI Foundation global observational study, ~30,000 participants), TRACK-HD (Tabrizi 2009 Lancet Neurology multi-center longitudinal study), and PREDICT-HD (Paulsen 2014 NIH pre-manifest study). Other anchors:

• HD Collaborative Research Group 1993 Cell — HTT gene discovery, CAG triplet repeat threshold (pathogenic >36).
• Shoulson-Fahn 1979 — Total Functional Capacity (TFC) 0-13 scale, HD staging framework (Stages 1-5).
• TRACK-HD (Tabrizi 2009 Lancet Neurology + Tabrizi 2013 Lancet Neurology) — multi-center longitudinal HD biomarker study; caudate atrophy rates 3-6%/yr.
• PREDICT-HD (Paulsen 2014 Neurology) — pre-manifest HD natural history cohort.
• ENROLL-HD (CHDI Foundation) — global observational study, current-standard HD natural history database.
• REGISTRY (European Huntington Disease Network) — predecessor European HD registry.
• Langbehn 2010 Clinical Genetics — CAG-age-of-onset prediction model (the Langbehn equation).
• Byrne 2018 Lancet Neurology — plasma NfL as HD prognostic biomarker.
• Wild 2015 / Byrne 2017 + Caron 2017 — single-molecule mHTT ELISA assay; target-engagement biomarker for HTT-lowering therapies.
• Aylward 2011 — caudate atrophy as HD imaging biomarker.
• GENERATION-HD1 (Tabrizi 2022 NEJM) — Tominersen Phase 3 trial (negative, redesigned for GENERATION-HD2).
• HD-CAB (Stout 2014) — HD Cognitive Assessment Battery.

---

Validation scorecard

The wrapper ships a 10-metric ENROLL-HD/TRACK-HD-anchored scorecard (validation_scorecard.json) that re-scores the dataset on every generation. Default seed 42 result:

ID	Metric	Target	Observed	Source
M01	CAG Repeat (Pathogenic)	40–48	44.14	HDCRG 1993 / ENROLL-HD
M02	UHDRS TMS — Early HD	23–47	36.10	TRACK-HD (Tabrizi 2009)
M03	UHDRS TMS — Late HD	75–115	98.89	Shoulson-Fahn / TRACK-HD
M04	TFC — Presymptomatic	12.2–13.0	12.97	Shoulson-Fahn 1979
M05	TFC — Late HD	0.5–4.5	2.77	Shoulson-Fahn 1979
M06	Caudate Volume — Late HD	1.0–4.0 mL	3.12	TRACK-HD (Aylward 2011)
M07	Caudate Atrophy %/yr	1.5–6.5%	3.29%	TRACK-HD / IMAGE-HD
M08	Plasma NfL — Early HD (pg/mL)	10–50	23.63	Byrne 2018 Lancet Neurology
M09	Family History 1st-Degree	0.75–0.95	0.840	HDSA / ENROLL-HD
M10	SDMT — Late HD	7–23	11.84	TRACK-HD / ENROLL-HD

Grade: A+ (100/100). Verified across seeds 42, 7, 123, 2024, 99, 1.

Standout calibration depth — this is a TRACK-HD/ENROLL-HD-grade dataset:

• M01 CAG 44.14 vs target 44 — 0.13 deviation 🎯
• M04 TFC Presymptomatic 12.97 vs Shoulson-Fahn 13 — 0.03 deviation 🎯
• M09 Family history 84% vs 85% — 1pp deviation 🎯
• TMS scores (M02 36.10, M03 98.89) reproduce the UHDRS clinical staging thresholds that define every HD clinical trial entry criterion
• Caudate atrophy 3.29%/yr matches TRACK-HD published rate within literature variance

---

Suggested use cases

• HTT-lowering ASO trial design simulation — placebo arm trajectory
  • treatment arm response modeling for Tominersen, branaplam, PRX-12-class therapies.
• mHTT target-engagement biomarker ML — plasma + CSF mHTT trajectory × treatment response × CAG count for ASO dose-finding.
• NfL prognostic biomarker validation — Byrne 2018 framework × longitudinal NfL trajectory × stage progression ML.
• CAG-driven progression modeling — Langbehn 2010 equation refinement + somatic instability × clinical onset prediction.
• Pre-manifest HD risk stratification — Presymptomatic CAG carriers
  • biomarker trajectory × prodromal conversion timing.
• Caudate atrophy ML for AI imaging vendors — TRACK-HD comparable caudate volumetric ML training.
• UHDRS cognitive battery composite scoring — SDMT + Stroop + Trails × multi-component cognitive decline.
• HD psychiatric phenotype ML — depression + apathy + irritability
  • suicidality × stage × CAG for behavioral pharmacology.
• Tetrabenazine / Deutetrabenazine response modeling — chorea reduction × patient features for VMAT2 inhibitor pharma.
• Tominersen Re-Trial Subgroup Enrichment — cognitively-defined responder identification for GENERATION-HD2-class trial design.

---

Loading

from datasets import load_dataset

ds = load_dataset(
    "xpertsystems/hcneu010-sample",
    data_files="HC_NEU_010_dataset.csv",
    split="train",
)

Or with pandas directly:

import pandas as pd
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="xpertsystems/hcneu010-sample",
    filename="HC_NEU_010_dataset.csv",
    repo_type="dataset",
)
df = pd.read_csv(path)

# Group by patient for longitudinal trajectory analysis
patients = df.groupby("patient_id")
for pid, sub in patients:
    tms_trajectory = sub.sort_values("visit_number")["uhdrs_total_motor_score"]
    cag = sub.iloc[0]["cag_repeat_allele1"]
    # ... fit progression by CAG repeat

The dataset ships with HC_NEU_010_schema.json providing per-column dtypes for pipeline integration:

import json
schema = json.load(open("HC_NEU_010_schema.json"))
# {"patient_id": "object", "cag_repeat_allele1": "int64", "uhdrs_total_motor_score": "float64", ...}

This dataset is longitudinal — multiple visits per patient, chronologically ordered. Visit cadence is semi-annual. For cross- sectional analysis, filter visit_number == 1 to get baseline rows only.

---

Schema highlights

Visit metadata & treatment — patient_id, site_id, visit_number, visit_date, years_from_baseline, disease_stage_at_visit ∈ {Presymptomatic, Prodromal, Early_HD, Middle_HD, Late_HD}, age_at_visit, sex, education_years, treatment_arm ∈ {Placebo, PDE10A_Inhibitor, HTT_ASO, Mitochondrial_Support}, treatment_adherence_pct.

Genetics — cag_repeat_allele1 (pathogenic, 36-75), cag_repeat_allele2 (wild-type), cap_score (CAG-age product), somatic_cag_instability, predicted_age_of_onset_years (Langbehn 2010 model), msh3_variant_flag, pms2_variant_flag (DNA repair modifier genes), family_history_hd_first_degree, de_novo_mutation_flag, genetic_testing_method.

UHDRS Motor — uhdrs_total_motor_score (TMS, 0-124), chorea_score (0-28), dystonia_score (0-20), bradykinesia_score, gait_score, tandem_walking_score, finger_tapping_dominant, pronate_supinate_dominant, rigidity_neck_score, dysarthria_score, dysphagia_score, diagnostic_confidence_level (1-4), tms_annual_progression_rate.

Cognitive — sdmt_score (Symbol Digit Modalities Test, primary HD cognitive measure), stroop_word_correct, stroop_color_correct, stroop_interference, trail_making_a_sec, trail_making_b_sec, verbal_fluency_letter, verbal_fluency_category, montreal_cognitive_assessment (MoCA), mini_mental_state_exam, cognitive_composite_score, annual_cognitive_decline_rate_sdmt.

Psychiatric (UHDRS-Behavioral) — uhdrs_behavioral_total, depression_score_phq9, anxiety_score_gad7, apathy_score (HD hallmark), irritability_score, obsessive_compulsive_score, psychosis_flag, suicidality_flag (HD-elevated risk), sleep_disorder_flag, psychiatric_hospitalization_flag.

Functional (TFC) — total_functional_capacity (0-13), tfc_stage (1-5), independence_scale (0-100), functional_assessment_scale, employment_status, living_situation, nursing_home_placement_flag, caregiver_burden_score, hd_quality_of_life_score.

Clinical — bmi, weight_change_kg_year, dysphagia_severity, falls_frequency_per_year, death_flag, cause_of_death ∈ {Aspiration, Cardiovascular, Pneumonia, Suicide}, study_dropout_flag, dropout_reason.

Imaging — caudate_volume_ml (HD primary atrophy marker), putamen_volume_ml, striatal_volume_ml, pallidum_volume_ml, whole_brain_volume_ml, cortical_thickness_frontal_mm, cortical_thickness_temporal_mm, white_matter_fa_corpus_callosum, white_matter_fa_cst, caudate_annual_atrophy_pct, mri_field_strength_T, mri_scanner_manufacturer.

Biomarkers — plasma_nfl_pg_ml, csf_nfl_pg_ml, plasma_mhtt_fg_ml (target-engagement for ASOs), csf_mhtt_fg_ml, mhtt_detected_flag, plasma_tau_pg_ml, plasma_gfap_pg_ml, csf_ykl40_ng_ml, il6_pg_ml, bdnf_pg_ml, nfl_annual_change_pct.

---

Calibration notes & limitations

In the spirit of honest synthetic data, a few things buyers of the sample should know:

1. Mean age at baseline 71.83 is above HD onset literature 30-50 years. The generator's age distribution is elderly-enriched and does not differentiate by stage (Presymptomatic should be younger than Late HD by ~20+ years). For age-stratified analysis, the full product calibrates age by stage per Langbehn 2010 / TRACK-HD.

2. Predicted age of onset = 80 years in this sample is implausible for CAG repeats 44.13 (Langbehn 2010 predicts onset ~45-55 years for CAG=44). The predicted_age_of_onset_years column appears to use a different prediction model than Langbehn; treat as relative rather than absolute prediction.

3. PHQ-9 ≥10 clinical depression rate 21% is below HD literature 40-50%. The generator's depression scoring is conservative; for HD psychiatric pharmacology research, the full product calibrates PHQ-9 distributions per ENROLL-HD published rates.

4. Nursing home placement only 1.5% is far below expected institutionalization rates over 8-year follow-up in Middle/Late HD (literature ~30-50%). Generator under-models institutionalization; for healthcare utilization modeling, use the full product.

5. HD-QoL mean 82.45 is preserved-quality across the cohort. Reflects high pre-symptomatic representation (30% Presymptomatic

   • 25% Prodromal + 25% Early HD = 80% of cohort with mild/moderate QoL impact). For symptomatic-only QoL analysis, filter to Middle
   • Late HD subsets.

6. CSF NfL mean ~2,100 pg/mL is at upper end of literature (Byrne 2018: CSF NfL ~500-3,000 pg/mL across HD stages). Acceptable but weighted toward symptomatic patients.

7. Treatment arm TMS progression rates show directionally-correct ordering (HTT_ASO 5.56 < PDE10A 5.77 < Placebo 6.72 < Mito 7.06) but compressed magnitudes vs generator's targeted effect sizes (target HTT_ASO -3.2 reduction; observed -1.16). The generator's treatment effects are attenuated; for trial-design modeling, the full product calibrates per published GENERATION-HD1, PRIDE-HD, LEGATO-HD trial outcomes.

8. CAG repeat distribution within stages is appropriately stratified — Presymptomatic 40.8, Prodromal 42.3, Early HD 45.9, Middle HD 46.6, Late HD 54.5. Clean monotonic increase matching CAG-stage relationship (higher CAG → earlier onset → later observation at higher stage).

9. mHTT detected flag 72% — reflects assay sensitivity limitations at low concentrations (Presymptomatic mHTT often below detection threshold). Clinically realistic.

10. Deterministic seeding. Wrapper passes user-specified seed through both np.random.default_rng() and np.random.seed(), and reassigns the generator's module-level rng for full reproducibility. Seed sweep verifies Grade A+ across {42, 7, 123, 2024, 99, 1}.

---

Commercial / full product

The full HC-NEU-010 product covers 5,000 patients × 16 semi-annual visits with refined Langbehn 2010 age-stratified cohort calibration, calibrated treatment effect sizes per GENERATION-HD1 / LEGATO-HD / PRIDE-HD outcomes, refined institutionalization modeling per ENROLL-HD real-world rates, PHQ-9 / GAD-7 / apathy scoring per CHDI HD-CAB published distributions, expanded biomarker panel (CSF p-tau, CSF neurogranin, plasma p-NfH), juvenile HD (JHD) cohort variant for CAG >60 modeling, REGISTRY-comparable cohort with European HD network demographics, pre-manifest gene-positive vs gene-negative case-control design, and HD-ISS (HD Integrated Staging System, Tabrizi 2022 Lancet Neurology) tagging. Available under commercial license — contact pradeep@xpertsystems.ai.

XpertSystems.ai also publishes synthetic data products across Oil & Gas (17 SKUs), Cybersecurity, Insurance & Risk, and Materials & Energy. Catalog: huggingface.co/xpertsystems.