Abstract:
Introducing a new drug to the market is time-consuming and costly. Therefore, reuse of existing drugs or drug combinations as therapeutics for diseases is identified to be much efficient and useful. This concept is known as drug repositioning/repurposing. The known drug combinations used for therapeutic effects are smaller than the total number of possible drug combinations. Hence, drug repositioning data for drug combinations naturally consists of a majority of unlabeled samples. Therefore, identifying reliable positives and reliable negatives is vital for a reliable binary classification model. With the assumption that unlabeled data is composed of both unidentified positive and negative samples, the set of unlabeled data has to be separated into positives and negatives by a reliable technique. In this study, the significance of employing Positive Unlabeled Learning (PUL) for drug combination repositioning is assessed.