E1.S3

DATASET DESCRIPTION

Disease Area

Pediatric Oncology — Hematologic Malignancies

Specific Research Focus

Genomic marker–driven leukemia subtype classification, MRD-based relapse prediction, and precision therapy matching in children and adolescents

Dataset Theme / Edition

Pediatric Leukemia AI Training Dataset, Cross-Sectional + Longitudinal Edition

Short Dataset Summary

A fully synthetic, biologically constrained pediatric leukemia dataset of 1,000 patients and 40,267 longitudinal clinical visits. Covers five diagnostic categories (ALL-B, ALL-T, AML, Mixed-lineage, No leukemia) with 41 cross-sectional features and 28 visit-level features per record, spanning demographics, hematology, genomics, and COG/BFM protocol-aligned treatment trajectories 

Intended Research Applications

Leukemia subtype classification, MRD-based relapse prediction, treatment response modeling, CAR-T and BMT eligibility prediction, genomic biomarker explainability (SHAP/LIME), pediatric survival analysis, TimeGAN conditional data generation, federated learning simulation across children's hospital networks 

Total Number of Patients

1,000 synthetic patients (90.2% leukemia-positive: ALL-B 545, AML 200, ALL-T 104, Mixed-lineage 53, No leukemia 98)

Number of Features / Parameters

69 unique features total — 41 cross-sectional (Part A) + 28 longitudinal visit-level (Part B); 20 static features + 19 normalized dynamic sequence features in ML-ready arrays

Data Modalities Included 

Demographics & predisposition, clinical presentation, hematology & bone marrow (CBC, blast%, LDH), genomics & molecular profiling (fusion genes, chromosomal abnormalities, somatic mutations, immunophenotype, MRD), risk stratification, longitudinal CBC trends, treatment response & toxicity, temporal event flags

Recommended AI Tasks

Multi-class classification, binary detection, regression (risk/survival scoring), sequence modeling (LSTM / Transformer), TimeGAN synthetic generation, survival analysis (Cox/AFT), XAI/SHAP feature attribution, federated learning 

Dataset Size

16.0 MB total across 9 files — CSV (Part A: 261 KB, Part B: 6.1 MB), NumPy padded array (1000 × 48 × 19, float32: 3.5 MB), attention mask, static/dynamic CSVs, and metadata JSON 

Release Date

May 2026 

Why This Dataset Exists

Real pediatric oncology data is extraordinarily difficult to access due to HIPAA/GDPR protections, pediatric consent requirements, and institutional data-sharing barriers. This dataset enables AI researchers, clinical data scientists, and startup teams to develop, benchmark, and validate leukemia ML models without requiring IRB approval or real patient data access

Real-World Problem Addressed

Each year ~400,000 children worldwide are diagnosed with cancer; leukemia is the most common. Late or misclassified diagnosis, failure to predict MRD-driven relapse, and suboptimal therapy matching are leading causes of treatment failure. This dataset directly supports AI development for earlier subtype identification, relapse risk stratification, and data-driven precision therapy selection in resource-constrained pediatric oncology settings