Mining Drug-Drug Interaction Induced Adverse Effects from Health Record Databases

Project Description

Major Research Products

Pattern Discovery from High-Order Drug-Drug Interaction Relations

Wen-Hao Chiang , Titus Schleyer, Li Shen, Lang Li, and Xia Ning. Pattern discovery from directional high-order drug-drug interaction relations. Journal of Healthcare Informatics Research, 2018. in press

Abstract: Drug-Drug Interactions (DDIs) and associated Adverse Drug Reactions (ADRs) represent a signicant public health problem in the United States. The research presented in this manuscript tackles the problems of representing, quantifying, discovering and visualizing patterns from high-order DDIs in a purely data-driven fashion within a unied graph-based framework and via unied convolution-based algorithms.We formulate the problem based on the notions of nondirectional DDI relations (DDI-nd’s) and directional DDI relations (DDI-d’s), and correspondingly developed weighted complete graphs and hyper-graphlets for their representation, respectively. We also develop a convolutional scheme and its stochastic algorithm SD2ID2 S to discover DDI based drug-drug similarities. Our experimental results demonstrate that such approaches can well capture the patterns of high-order DDIs.

DDI Prediction based on Co-Medication Patterns and Graph Matching

Wen-Hao Chiang , Li Shen, Lang Li, and Xia Ning . Drug-drug interaction prediction based on co-medication patterns and graph matching. Short version in International Conference on Intelligent Biology and Medicine, ICIBM’18. Long version under review in Int. J. Computational Biology and Drug Design.

Abstract: High-order Drug-Drug Interactions (DDIs) and associated Adverse Drug Reactions (ADRs) are common, particularly for elderly people, and therefore represent a significant public health problem. Currently, high-order DDI detection primarily relies on the spontaneous reporting of ADR events. However, proactive prediction of unknown DDIs and their ADRs has indispensable benefit for protective health care. In this manuscript, the problem of predicting whether a drug combination of arbitrary orders is likely to induce adverse drug reactions is considered. The prediction problem becomes highly non-trivial when arbitrary orders of drug combinations have to been accommodated by the prospective computational methods. To solve this problem, novel kernels over drug combinations of arbitrary orders are developed within support vector machines for the prediction. Graph matching methods are used in the novel kernels to measure the similarities among drug combinations, in which drug co-medication patterns are leveraged to measure single drug similarities. The experimental results on a real-world dataset demonstrated that the new kernels achieve an area under the curve (AUC) value 0.912 for the prediction problem. The new methods with drug co-medication based single drug similarities can accurately predict whether a drug combination is likely to induce adverse drug reactions of interest.