SP1 - COMMITMENT

Joint infrastructure and project management

A fundamental requirement for novel systems-medicine approaches is the leveraging of massive data resources to improve generalizability of algorithms and demonstrate biological reproducibility. This faces the substantial challenge that data sets can frequently not be combined in a single storage solution, due to consent, legal and logistic restrictions. To address this, COMMITMENT’s subproject “SP1 – Joint infrastructure and project management” will implement an innovative computational infrastructure for privacy-preserving, distributed transfer machine learning. This infrastructure will allow clinical and biological data to remain stored at their source locations and algorithms to be optimized in the feature space without compromising personally identifying information. This will provide the basis for integration of mechanistic knowledge defined as part of SP3, the development of advanced transfer-learning procedures as part of SP4, and the integrative application of these tools on patient and large scale lifespan data during SP5 and SP6. It will further facilitate the application of COMMITMENT algorithms across data resources developed in SP2, for algorithm validation and testing of associations with clinical outcome. A particular focus of this subproject is making the infrastructure system publicly available to the scientific community and to facilitate a new generation of data exchange, based on privacy- protected machine-learning in existing legacy datasets. This subproject will further be concerned with the project and scientific coordination of COMMITMENT, contractual and communication management, (within COMMITMENT, the wider scientific community and the general public), multi-disciplinary training of PhD and postdoctoral scientists, and the protection of potential intellectual property arising during the project. Furthermore, we will rigorously review all analyses steps performed during COMMITMENT, to safeguard highest scientific quality standards.