Drug-drug interactions (DDIs), which can add to or diminish the effect of one drug or impact the metabolism of one drug, have harmful effects on health in patients that take multiple drugs. Testing for DDIs is slow and costly, so computational models have recently been used to predict them. Network information is useful in describing a drug’s known interactions and mechanisms to determine whether two drugs could be interacting. This research explores various machine learning models to predict the severity of unknown interactions of existing drugs (major, minor, or moderate), using the DDInter database.
Drug-drug interactions (DDIs), which can add to or diminish the effect of one drug or impact the metabolism of one drug, have harmful effects on health in patients that take multiple drugs. Testing for DDIs is slow and costly, so computational models have recently been used to predict them. Network information is useful in describing a drug’s known interactions and mechanisms to determine whether two drugs could be interacting. This research explores various machine learning models to predict the severity of unknown interactions of existing drugs (major, minor, or moderate), using the DDInter database. The best-performing model, a multi-layer perceptron model which concatenates two drugs’ embeddings, had an accuracy of about 92%, average precision of about 87.1% (89.4% for the major category, 78.5% for minor, and 93.5% for moderate) and average recall of 83.3% (86.4% for major, 70.1% for minor, and 95.3% for moderate). Future work in this area includes refining models to increase precision and recall, obtaining a more balanced dataset, predicting the mode of interaction, and using other drug characteristics. The code for this project can be found here, along with a Streamlit deployment of the model: https://github.com/deetyabn/DDI_severity_prediction.
Explore More!
Related Projects
© 2022 Inspirit AI. All Rights Reserved.