A Machine Learning and Network Framework to Discover New Indications for Small Molecules

In 2020, researchers developed a machine learning model named CATNIP to predict new uses for existing drugs, a process known as drug repurposing. The model was trained using 2,576 small molecules and 16 drug similarity features, achieving significant predictive power with an area-under-the-receiver-operator curve (AUC) of 0.841. The study found that drugs do not necessarily have to target the same genes, but rather the same biological pathway, to share a clinical indication. The model identified potential new treatments for neurological diseases and diabetes, including the repurposing of antidepressants for Parkinson's disease and a thyroid cancer drug for diabetes. However, the model had limitations such as data availability and its application to rare diseases.