Algorithm for Predicting Bitterness of Children’s Medication

The taste of drugs has always been a key factor affecting children’s compliance. The children’s preference for bitterness will cause them to fail to take the medicine on time and affect the treatment effect. To understand the taste of drugs is the premise of the study of drug correction and masking. The previous research methods usually use the oral taste method or the electronic tongue method to judge the taste of drugs, which has some inconveniences. In this paper, the smiles of the compound were used to derive the physicochemical descriptors and the MACCS fingerprint, and the prediction model of the bitter taste of children’s drugs was constructed based on the random forest algorithm and XgBoost algorithm. The hospital data were used for verification and analysis to provide help for the development of correction and masking of children’s medicines. In this study, we first collected the bitter and non-bitter data by referring to the literature, using smiles of compounds to calculate the physicochemical descriptors and MACCS fingerprints as features, and using Pearson coefficients and SVM-REF for feature dimensionality reduction. Then, based on the random forest algorithm and XgBoost algorithm, the bitterness prediction model for children was constructed, and the 10-fold cross validation was used to optimize the model. Finally, the model is applied to some datasets to predict the content of bitter compounds. The results show that MACCS fingerprint is the most suitable for the prediction of bitter compounds. The classification effect of XgBoost is better than that of random forest. Model of xgboost-maccs has the best classification effect with 88% accuracy, and model of rf-descriptors has the classification effect with 83% accuracy.