Drug clearance pathway prediction based on semi-supervised learning

It is necessary to confirm that a new drug can be appropriately cleared from the human body. However, checking the clearance pathway of a drug in the human body requires clinical trials, and therefore requires large cost. Thus, computational methods for drug clearance pathway prediction have been studied. The proposed prediction methods developed previously were based on a supervised learning algorithm, which requires clearance pathway information for all drugs in a training set as input labels. However, these data are often insufficient because of the high cost of their acquisition. In this paper, we propose a new drug clearance pathway prediction method based on semisupervised learning, which can use not only labeled data but also unlabeled data. We evaluated the effectiveness of our method, focusing on the cytochrome P450 2C19 enzyme, which is involved in one of the major clearance pathways.