IHPreten: A novel supervised learning framework with attribute regularization for prediction of incompatible herb pair in traditional Chinese medicine

Abstract Adverse drug-drug interaction is a critical safety issue for the development of drugs. In Traditional Chinese Medicine (TCM), adverse herb-herb interaction is regarded as negative reactions in patients after the absorption of the decoction of Incompatible Herb Pair (IHP). Recently, many methods are proposed for IHP researches, but most of them focus on revealing and analyzing the adverse reactions of known IHPs, despite that there are still a number of new IHPs discovered by accidents. Up to now, IHPs have become a serious threat to public health in TCM medication. In this paper, we propose a novel supervised learning framework with attribute regularization for IHP prediction. In this framework, we model the prediction task as a non-negative matrix tri-factorization problem, in which two important herb attributes (efficacy and flavor) and their correlation are incorporated to characterize the incompatible relationship between herbs. A hypothetical test method is adopted to evaluate the statistical significance of the dissimilar characteristics of two attributes and the attribute information from the TCM literature is adopted to estimate the correlation between attributes. These two constraints are jointly incorporated as attribute regularizations into the framework to improve IHP prediction. The update solutions and the convergence proof for the optimization problem are given in detail. Experimental results on the real-world IHP datasets demonstrate that the proposed framework is effective for IHP prediction compared with eight baseline methods and its variants.