E1.L2

DATASET DESCRIPTION

Disease Area

 Oncology — Breast Cancer 

Specific Research Focus

 Early detection of distant metastatic spread in breast cancer patients; identifying the transition from localized to metastatic disease before clinical presentation 

Dataset Theme / Edition

Synthetic Early Metastatic Breast Cancer Detection Dataset 

Short Dataset Summary

 A large-scale, fully synthetic, clinically-grounded tabular dataset of 50,000 breast cancer patient records spanning 91 features across 14 clinical domains, from genomics and liquid biopsy biomarkers to imaging characteristics, symptoms, treatment history, and longitudinal monitoring. Biologically coherent subtype-specific correlations are enforced throughout. Primary target is binary metastasis detection at 17.5% positive-class prevalence. 

Intended Research Applications

 Early metastasis classification · Survival analysis (Kaplan-Meier, Cox PH) · Recurrence risk stratification · Treatment response prediction · SHAP / XAI feature importance · Multi-task learning · Federated learning benchmarking · Fairness and bias auditing · Uncertainty quantification · Genomic-clinical multi-modal fusion 

Total Number of Patients

50,000 synthetic patient records

Number of Features / Parameters

91 features** (excl. Patient_ID) across 14 clinical domains + 6 target/secondary target columns 

Data Modalities Included

 Demographics & Risk · Germline Genomics (BRCA1/2, TP53, TMB, CNV, RNA signature) · Tumor Pathology (size, grade, receptors, Ki-67, LVI) · Imaging proxies (BI-RADS, MRI, Ultrasound, Calcifications) · Liquid Biopsy (ctDNA, CTC count) · Standard Labs (CBC, liver panel, CRP, ESR) · Symptomology (8 clinical symptoms) · Longitudinal monitoring (biomarker trends, imaging progression) · Treatment history · Metastasis site likelihoods 

Recommended AI Tasks

 Binary classification · Multi-label classification (metastasis site) · Survival / time-to-event modeling · Regression (risk scores, survival probabilities) · Imputation benchmarking · Federated learning · Explainability / SHAP analysis · Calibration studies 

Dataset Size

 CSV: 22.5 MB · Compressed GZ: 4.9 MB · In-memory (pandas): ~78 MB 

Release Date 

May 2026

Why This Dataset Exists

 No large-scale, richly annotated, publicly shareable breast cancer dataset exists that combines genomics, liquid biopsy, imaging proxies, and longitudinal biomarker trends in a single privacy-safe, reproducible format. Real patient-level datasets of this scope are locked behind institutional access agreements, making algorithm development and benchmarking slow and inequitable. SynthOnco-BC50K provides a fully open, clinically realistic alternative. 

Real-World Problem Addressed

 ~30% of early-stage breast cancer patients will eventually develop metastatic disease, yet most are not identified until symptoms appear — often years after the window for preventive intervention has closed. AI models that can flag high-risk patients early using routine clinical data (labs, imaging reports, ctDNA) could enable proactive monitoring and earlier therapeutic escalation, directly impacting the ~500,000 breast cancer deaths globally each year. This dataset exists to make building and testing those models accessible to any research team, anywhere.