The initial idea of VANTAGE6 came from the Personal Health Train (PHT). The PHT is a concept in which partitioned datasets (hosted and maintained by different organizations) can be used to compute a global model without combining the partitioned dataset. It does so by computing partial models which are combined at a later stage. This way, the privacy of the individual patients is secured, as aggregated statistics (i.e., the partial model) are shared rather than raw patient records.
However, VANTAGE6 has also other potential applications.
- FAIR data-point
An alternative application of VANTAGE is to create a FAIR data-point. This point can provide an API interface to their data and allowing them to control which algorithms are allowed to run.
In this case only, a single node is used. Researchers can ask questions to this node within the limits set by the data-provider (which hosts the node). This way, researchers are able to execute complex algorithms without compromising the privacy of the individual patients.
- Model repository
Anoter application of VANTAGE6 is functioning as a pre-computed model repository. This is different from federated learning, since the model is already computed, so it no longer needs access to the data. That being said, it is possible to train this model on a partitioned dataset using VANTAGE6 first. After training, the model can be stored in the repository and used by the researchers. VANTAGE6 provides an API to these models, allowing other application (or researchers) to use these models.