Drug interaction prediction framework and prediction method based on multi-modal representation learning

The invention discloses a drug interaction prediction framework and prediction method based on multi-modal representation learning, and the prediction framework comprises the steps: extracting drug features through the multi-modal representation learning, connecting a feature vector of a small-molecule drug with a feature vector of a biological pharmacy to form a drug pair vector, and inputting each feature in the drug pair vector into a neural network for operation, and averaging the operation results of all drug pair features. The prediction method comprises the steps of firstly constructing a sample, inputting the sample into a prediction framework for training and testing, and finally inputting a drug pair for prediction. According to the method, a two-channel CNN is provided to effectively extract complex local chemical information and context relationships in a sequence, association information of all drug nodes in a heterogeneous network is coded into one-dimensional feature vectors, finally, DNN is us