Method Research For The Prediction Of Drug’s Side Effect Based On Information Integration

Drug side effect prediction is an important content of the new developing drugs’ safety evaluation, which is directly related to the health of human beings. Predicting side effect in time and accurately is of vital significance to reduce the incidence of adverse drug events and to protect the health of patients. Although in vitro safety experiment can be used to predict side effects of drugs, it remains very challenging in terms of cost and efficiency. Therefore, the need of computational methods for predicting drugs’ side effects is growing. In this thesis, we selected biological molecules as the research objects. By mining different kinds of relating knowledge, we predict the associations between biological molecules and side effects. The associations can help to provide reference for clinical use of drugs. The main contents are as follows:1. We predicted the interactions between drugs and side effects based on drugs’ chemical substructure using improved BP Neural Networks. We use drugs’ chemical substructures as features with target protein and functional group as comparision. By mining potential information from drug-features-side effects network, we predicted drugs’ side effects. Results show that the BP Neural Network model is suitable and effective for side effect prediction. It can excavate some potential associations between side effects and chemical sub-structures.2. We predicted the interactions between drugs and side effects based on multi-label ensemble K nearest neighbor via multiple feature fusion. In detail, we select drugs chemical property, biological property, phenotypic property and pharmacological property as features. By optimizing the weight of single models in the ensemble learning model, the optimal solution is obtained. Experimental results show that, our method has high precision and accuracy score as well as good robustness, and can successfully predict some side effects, which would not show until drug was sent to the market.3. We implemented a visual side effect prediction tool based on three algorithms named DSEP. The tool can provide two kinds of services:database service and molecular file service. Users can use this tool through Client mode which supports in batch side-effects prediction or web server mode which can conduct on-line calculation.