Prediction of Side Effects Using Comprehensive Similarity Measures

Figure 1

System overview. (a) To build a set of features for a drug-side effect pair, the maximum similarity was selected for each feature based on the known associations from the training samples between the side effect and other drugs. (b) At this step, the maximum side effect anatomical hierarchy similarity was chosen based on the known side effects of the drug from the training samples. (c) By assigning values, as was done for (a) and (b), datasets for machine learning were created with different combinations of features, and diverse classification algorithms, including a random forest, XGBoost, logistic regression, and naive Bayesian model, were applied to predict the relationship between a side effect and a drug. (d) Stacking ensemble learning that incorporated all four classifiers as its base classifiers and used a neural network as its meta classifier was applied.

(a)

(b)

(c)

(d)