Artificial intelligence approaches for privacy-protected pooling of genomics, clinical and other ‘-Omics’ data analysis for haematological disease prognosis, prevention, diagnostics and treatment

There are up to 450 Hematological Diseases (HDs), generally classified in six large groups of oncological and non-oncological diseases. HDs result from ab-normalities of blood cells; lymphoid organs; and coagulation factors, and affect a substantial number of patients. For example, HDs account for about 5% of Cancers. Most HDs can cause chronic health problems and many of them are life-threatening conditions requiring numerous resources for correct diagnosis, management and treatment. Recently, the European Hematology Association (EHA) evaluated the financial burden of blood disorders on European society at €22.5 billion per year. Personalized or precision medicine is a medical model in which conventional medicine is combined with advanced genetic profiling, leveraging Artificial In-telligence (AI) and Machine Learning (ML). This results in tailored diagnostic, prognostic and therapeutic strategies. Personalized medicine can revolutionize hematology, improving patients’ quality of life and reducing the overall financial burden. Unfortunately, this is currently underexplored: existing AI models for HDs lack patient-centricity and personalization. Furthermore, the vast amounts of relevant data produced are often inaccessible and diffused. In this thesis, we support the pooling of genomic, clinical data and other “-omics” health data for precision medicine in hematology. To do so, we Collect, compatibilize, and prepare relevant data from over ten clinical repositories. We then leverage this data to validate existing prognostic models for HDs at scale. Subsequently, we develop novel precision medicine AI and ML models for prog-nosis in hematology. Finally, we show that such models can be securely trained and integrated for analysis and application, without breaching patients’ rights and trust. This is achieved by leveraging interoperability standards and novel privacy enhancing technologies (PETs) such as federated learning and synthetic data.