Genome-Centric Multimodal Data Integration in Personalised Cardiovascular Medicine.


The Swiss AI Lab IDSIA (Istituto Dalle Molle di Studi sull’Intelligenza Artificiale) is a not-for-profit research institute for artificial intelligence. IDSIA was founded in Lugano in 1988 by Angelo Dalle Molle (1908-2002), an Italian philanthropist whose vision was a world where technological progress and human development could both contribute to the improvement of our quality of life.
Since the foundation of USI and SUPSI in year 2000, IDSIA has been affiliated to the two Universities and designated to be a “bridge” between them. Therefore, its aim is to be active in both basic and applied research. This purpose is achieved by developing innovative ideas and algorithms in basic research projects and transferring them into real world applications, in partnership with the industry. We focus on various aspects of artificial intelligence, from deep neural networks to explainable causal networks, from machine vision to autonomous robotics, intersecting control engineering and operations research with AI/ML methods.
IDSIA is also committed to transfer its knowledge by teaching and mentoring students at the bachelor, master, and PhD level. Finally, IDSIA aims at contributing to the current dialogue with the society about ethical artificial intelligence, in particular issues related to the transparency and fairness of machine learning algorithms.
IDSIA is situated in Lugano, a lakeside city in the Italian-speaking canton of Ticino, a region of Switzerland well known for its warm climate and outstanding scenery.

Federated learning is an effective strategy for learning from distributed data without moving them to a central site. This, combined with privacy-preserving methods (differential privacy, homomorphic encryption, Multi-Party Computation, etc.) allows learning from bigger datasets while respecting the strict privacy requirement necessary to the sensitive data involved in medical research. Many recent applications have proven the feasibility of the federated approach to machine learning and have led to the development of effective methods for its implementation. However, in the field of genomics these methods are in their infancy and stable tools for federated analysis are still under development. In the NextGen project, IDSIA will work to the development and integration in the Pathfinder of federated machine learning and genomic data analysis methods, including polygenic risk scores, clustering and dimensionality reduction, supervised learning, and deep learning. Furthermore, as privacy protection is the main purpose for employing federated learning in this project, but at the same time, federated learning alone does not ensure it, we will work to guarantee robustness of the developed tools with respect to privacy threats. Finally, we aim to leverage our expertise in AI to integrate multi-modal data for modeling cardiovascular diseases and to improve and (semi-)automatize genomic data curation, integration, and interpretation pipelines which are still often time-consuming and sub-optimal manual processes.