A Predictive Modeling Study Of Drug-drug Interactions Based On Drug Transcriptome Data

Drug-drug interactions?DDIs?are the compounding effects of two or more drugs taken simultaneously over a period of time.In the course of a doctor's treatment of a patient,multiple medications are often taken to achieve a certain therapeutic effect,which also puts the patient at risk of drug interactions.Studies in recent years have shown that DDIs are consuming significant public health resources,with billions of dollars spent annually on treating toxic side effects caused by DDIs.The discovery of DDIs poses a huge challenge as biological experiments to discover drug-drug interactions require a significant investment of human and material resources.Building accurate predictive models for DDIs has become a hot topic in drug research.With the development of high-throughput sequencing technology in recent years,multiple drug-induced transcriptome data have been accumulated in the LINCS L1000 database,which provides new ways of characterizing drugs and new ideas for building predictive models for DDIs.This study addresses two major issues in constructing more accurate predictive models for DDIs based on drug transcriptome data.1)Machine noise in drug transcriptome data,for which this study uses the drug structure information and the similarity relationship between the drug structures to embed the drug transcriptome data through the graph convolution neural network.The embedded transcriptome data were found to be more useful for the prediction of DDIs.2)We explored whether treating drugs as sequence information is more conducive to the prediction of DDIs,thus introducing the Long Short Term Memory networks?LSTM?model for the prediction of DDIs,and finally proposed the GEDDI?graph embedding for drug-drug interaction prediction?method,which predicts the DDIs of drugs based on drug transcriptome data and achieves F1-score_marco of up to 95.3%,significantly higher than other machine learning methods.Type 2 diabetes is a common endocrine metabolic disease and has become one of the major diseases that threaten human health.Since type 2 diabetes is often associated with multiple complications and requires the use of multiple medications,the probability of DDIs is higher.In recent years multiple glucose-lowering drugs have been used in the clinic and it is necessary to explore the interactions between these drugs and other drugs.This study used GEDDI to explore the interactions between common type 2 diabetes medications and between type 2 diabetes medications and other chronic disease medications,and found that the combination of multiple medications and diabetes medications resulted in multiple adverse effects such as hypoglycemia,lactic acidosis,angioedma,and others.The results of this study will provide an important reference for drug-related research based on drug transcriptome data and provide reasonable guidance for medication management in patients with diabetes.