Calming the Storm: Identifying Multi-Cytokine Inhibiting Drugs with Machine Learning for COVID-19 Induced Cytokine Storms

In COVID-19, patients in severe condition often suffer from a major complication that leads to lung injury, ARDS and possibly death: the cytokine storm. The cytokine storm is composed of many cytokines, including IL-6, IL2, and TNF-? for COVID-19. To combat such effects, a cocktail of cytokine-inhibiting drugs are administered. However, a combination of drugs can be overly taxing upon the patient, thus creating the demand for a drug that targets multiple cytokines. This project identifies multi-cytokine inhibiting compounds from FDA-approved drugs with machine-learning methods. Many machine-learning algorithms were applied to the task and Support Vector Machines proved best with strong performances across all cytokines. Under the constraints of limited data (30–60 samples) for some cytokines, we significantly boosted modeling power and accuracy with the application of data dimension reduction technique, Principle Component Analysis. After exhaustive exploration, the FDAapproved hepatitis-C drug—glecaprevir—was identified with confidences of 80.52% for TNF-?, 99.04% for IL-2, and 98.23% for IL-6.