A research team at KAIST has further extended our understanding of drug-drug interaction (DDI) and drug-food interaction (DFI) with the development of a computational framework called DeepDDI. The effects of drugs can be influenced by other drugs or food constituents that are consumed together. These interactions may also incur adverse side effects. Hence, understanding of DDI and DFI is crucial in ensuring the predictability of the effects of a drug.
The relevancy of the study stems from the high percentage of patients taking five or more medications and the fact that side effects due to DDIs are one of the leading causes of drug withdrawals from the market. However, analysis of these interactions has been hindered by the logistical difficulties in conducting experimental studies; it is almost impossible to account for the different types of food the patient consumes, made even more difficult by the time-consuming and costly nature of the experiments.
As a result, researchers have turned to computational approaches to the issue. Previous methods have been limited by the lack of DDI information with regards to the pharmacological effects and the requirement of comprehensive details of the drugs — which are often unavailable — to predict DDIs. Therefore, these previous approaches were mainly applicable to drugs with known mechanisms.
The research team led by Dr. Jae Yong Ryu, Assistant Professor Hyun Uk Kim, and Distinguished Professor Sang Yup Lee from the Department of Chemical and Biomolecular Engineering has tackled the issue and presented a computational framework — DeepDDI. DeepDDI takes only the names and structural information of the drugs as inputs and predicts the appropriate DDIs for the drugs.
According to the study published in the Proceedings of the National Academy of Sciences of the United States of America, this new framework can generate 86 types of interactions using a deep neural network with a mean accuracy of 92.4%. More importantly, the output of DeepDDI is in the form of human-readable sentences that describe the risks of adverse drug events and possible changes in the pharmacological effects of the drug. For example, the input of two drugs — oxycodone and atazanavir — generated, “The metabolism of Oxycodone can be decreased when combined with Atazanavir … the risk or severity of adverse effects can be increased when Oxycodone is combined with Atazanavir.”
The applications of this new development is not limited to predicting interactions and side effects; it can also suggest alternative drugs that can yield the desired pharmacological effects without unintended consequences, as well as a list of drugs and food to avoid in order to achieve maximum efficacy of the prescribed drug. Also, the effects of food constituents and their bioactivities can be analyzed using DeepDDI.
Regarding the implications of the study, Professor Lee commented, “We have developed a platform that will allow precision medicine in the era of the Fourth Industrial Revolution. DeepDDI can provide crucial information on drug prescription and dietary suggestions to maximize health benefits and ultimately help maintain a healthy life in this aging society.”