Projects

The Korean COVID-19 Multi-Omics (KCMO) projects aims to track chagnes in immune responses according to the disease progression of patients using COVID-19 longitudinal multi-omics data.

  • Development and validation of a deep learning-based severity prediction model using longitudinal clinical and single-cell data of COVID-19 patients

    LED BY

      TBD

    DESCRIPTION

    This study focuses on using longitudinal clinical data to predict the need for oxygen support in COVID-19 patients up to three days in advance, aiming to identify those with deteriorating health conditions early. A deep learning GRU (Gated Recurrent Unit) model, trained on clinical data from 4,454 patients across multiple South Korean hospitals, demonstrated up to 93% accuracy in predicting the risk of requiring critical care. Validation on an external dataset of 444 patients showed 88% accuracy, and testing on non-COVID pneumonia patients yielded 73.8% accuracy, suggesting the model's potential applicability to broader respiratory infectious diseases. This predictive model could aid in the efficient allocation of isolation beds and resources, preparing for future infectious disease pandemics.