Method for Predicting Compound Properties in Drug Development Based on Machine Learning

Data on drugs, targets and indications were collected from the database, and gene expression profile data processed by 1309 small drug molecules were collected from the connectivity map. The biclustering algorithm was used to cluster 1309 small drug molecules in CMap, and the intersection of CMap drugs and drugs in the three databases of DrugBank, TTD and DGIdb was taken. The PubChem tool was used to calculate the chemical structure similarity of the drug, and then the Tanimoto coefficient between the two drugs was calculated from the collected drug data. Using Tanimoto coefficients as parameters, support vector machines, naive Bayes and logistic regression were used to construct combined drug prediction models under different positive and negative sample ratios, and the support vector machines were selected as the best model. Use the nearest neighbor recommendation algorithm to build five similarity models and use logistic regression as the integrated learning algorithm to build the integrated model and perform feature screening. The AUROC value of the integrated model constructed using drug target similarity, drug indication similarity, drug chemical structure similarity and drug expression profile similarity was 0.89, and the AUPR value was 0.383. Applying the optimal model to the prediction of paclitaxel combination medication can draw the performance prediction of the combination medication. Preliminary experiments have proven that the predicted drug combination has a better combined synergistic effect.