Prediction of Side Effects Using Comprehensive Similarity Measures

Biomed Res Int. 2020 Feb 27:2020:1357630. doi: 10.1155/2020/1357630. eCollection 2020.


Identifying the potential side effects of drugs is crucial in clinical trials in the pharmaceutical industry. The existing side effect prediction methods mainly focus on the chemical and biological properties of drugs. This study proposes a method that uses diverse information such as drug-drug interactions from DrugBank, drug-drug interactions from network, single nucleotide polymorphisms, and side effect anatomical hierarchy as well as chemical structures, indications, and targets. The proposed method is based on the assumption that properties used in drug repositioning studies could be utilized to predict side effects because the phenotypic expression of a side effect is similar to that of the disease. The prediction results using the proposed method showed a 3.5% improvement in the area under the curve (AUC) over that obtained when only chemical, indication, and target features were used. The random forest model delivered outstanding results for all combinations of feature types. Finally, after identifying candidate side effects of drugs using the proposed method, the following four popular drugs were discussed: (1) dasatinib, (2) sitagliptin, (3) vorinostat, and (4) clonidine.

Publication types

  • Comparative Study
  • Evaluation Study

MeSH terms

  • Area Under Curve
  • Clonidine / pharmacology*
  • Dasatinib / pharmacokinetics*
  • Drug Interactions
  • Drug Repositioning
  • Drug-Related Side Effects and Adverse Reactions
  • Humans
  • Machine Learning*
  • Polymorphism, Single Nucleotide / genetics
  • Sitagliptin Phosphate / pharmacology*
  • Vorinostat / pharmacology*


  • Vorinostat
  • Clonidine
  • Dasatinib
  • Sitagliptin Phosphate