Synthema offers a pan‑European, privacy‑by‑design platform that uses federated learning, secure multi‑party computation and differential privacy to train AI models on local hematology data without moving raw patient records. The trained models generate high‑quality synthetic multimodal datasets—clinical, omics and imaging—that replicate real‑world patterns and can be freely shared for research, drug development and clinical decision support in rare blood disorders.
Funding
Funding not disclosed
Founders
Product
Problem
Rare hematological diseases suffer from limited patient numbers and fragmented data across hospitals and registries, making it difficult to obtain the sample sizes needed for robust research and precision‑medicine development. The lack of interoperable, privacy‑compliant data sharing hampers the creation of reliable diagnostic models and treatment‑outcome studies.
Solution
Synthema provides a pan‑European, privacy‑by‑design platform that connects clinical sites through federated learning, allowing AI models to be trained on local data without transferring raw patient records. The infrastructure incorporates secure multi‑party computation and differential privacy to protect individual information while enabling collaborative model development. Trained models generate high‑quality synthetic multimodal datasets—including clinical, omics, and imaging data—that preserve the statistical properties of real patients, effectively creating “virtual patients.” These synthetic datasets can be freely shared for research, drug development, and clinical decision‑support without breaching GDPR or other data‑protection regulations. By aggregating synthetic data across borders, Synthema expands the usable data pool for rare blood disorders such as sickle cell disease and acute myeloid leukaemia.
Target Audience
Primary users are academic and industry researchers, pharmaceutical developers, and healthcare organizations focused on rare hematological diseases who require large, privacy‑safe datasets for AI‑driven studies.
Features
- Federated learning network that trains AI models on-site, eliminating the need to exchange raw health data
- Secure multi‑party computation (SMPC) for privacy‑preserving model aggregation across participating institutions
- Differential privacy mechanisms that enforce strict limits on information leakage from each clinical site
- Synthetic data generation engine producing multimodal virtual patient records (clinical, genomics, imaging) that retain real‑world patterns
- GDPR‑compliant architecture aligned with the European Health Data Space framework
- Open‑science collaboration tools and multidisciplinary training resources for researchers and clinicians