Requirements Federated Learning and mUlti-party computation Techniques for prostatE cancer, published by HL7 Europe. This guide is not an authorized publication; it is the continuous build for version 0.1.0 built by the FHIR (HL7® FHIR® Standard) CI Build. This version is based on the current content of https://github.com/hl7-eu/flute-requirements/ and changes regularly. See the Directory of published versions

4.2 Requirements overview

The user interface for the researcher/innovator will be a jupyterlab environment in which they can query the federated data (the union of data sets at the data owners’ premises) in a privacy-preserving way, i.e., without the possibility to reveal sensitive information. In the following, we will call queries issued by the researcher/innovator “researcher queries”. In particular, the interface will allow the user to choose researcher queries from a list of researcher queries for which federated algorithms have been implemented in the platform (or compositions of such queries).

The FLUTE platform will be an extension of the TRUMPET platform and hence should support all the researcher queries supported by TRUMPET. Moreover, the FLUTE platform should at least support the following researcher queries with federated, privacy-preserving implementations (more details are provided in Section 4.3):

• For prediction tasks
  • Logistic Regression (LR)
  • Decision Trees (DT) and Random Forests (RF)
  • Support Vector Machines (SVM)
  • Deep Neural Networks (DNNs), including Convolutional Neural Networks (CNNs)
  • Meta-Algorithms: Bayesian Optimization, Grid Search…
• For synthetic data generation tasks
  • Generative adversarial networks (GAN)
  • Variational auto-encoders (VAE)
  • Diffusion models

The interface of the platform to the data hub owners should be a secure software component, accessing sensitive data at the data owner’s premises and only disclosing to the outside world, using appropriate multi-party computation (SMPC), information which may be disclosed according to the data owner’s constraints.

The data owner node should provide the following interfaces:

• An interface to the local database, providing HL7-FHIR compliant data access
• A server which can be contacted by the platform, but is also sufficiently secured
• A user interface for the data owner, which allows the data owner to specify
  • The provided data, with for each set of data possibly privacy requirements and other details
  • Other platform users (data owners and researchers) and the extent to which they are trusted
  • Study agreements, i.e., agreements describing studies on which the data owner agreed with other data owners and researcher, and the details of such studies
  • Administration, e.g., secret keys as they are needed for the protocols researched in WP2.