Prediction of Drug-Target Interactions Based on Multi-Modal Autoencoder

Prediction of drug-target interactions (DTIs) plays a crucial role in various areas of drug development, such as drug reuse and identification of potential side effects of drugs. In recent years, although great progress has been made in DTIs prediction, the existing methods still have the problem of high sparsity of drug-target datasets. In this work, the multi-modal autoencoder is used to fuse the similarity networks of many different drugs (targets), so as to extract the low-dimensional feature vectors of drugs (targets), and input them into the knowledge graph embedding (KGE) and neural factorization machine (NFM) models proposed by our predecessor to calculate the drug-target interaction prediction score. In this regard, the method can also achieve accurate and robust predictions on the gold standard dataset.