Prediction of long-/post-COVID by artificial intelligence using a federated learning approach on German cohort data

The emergence of post-COVID syndrome in people with previous SARS-CoV-2 infection has been increasingly observed. Despite the increasing number of cases, a lack of mechanistic understanding of this disease has led to difficulties in diagnosing and providing relevant medical care to the patients.

During the COVID-19 pandemic, tremendous efforts across the world in the fight against infection led to generations of vast amounts of longitudinal clinical and research data sets from diverse populations. In this project, we propose to use federated learning to combine the power of these vast research data sets across three different cohorts along with big data analysis approaches to investigate in the disease pathogenesis of post-/long-COVID phenotypes. To achieve this goal, the project will be divided into three main subprojects. Subproject one will be focused on identifying the differences in the data structure across three different cohorts and providing combined accessibility of these data sets. Subproject two will focus on harmonization of various multi-OMICs measurements generated across these cohorts along with analysis model construction. Finally, subproject three brings together the outputs of first two subprojects by developing federated machine learning approaches for analysis of these big data sets, geared towards long-/post-COVID phenotypes.