Prediction of side effects of polypragmation using a graphical neural network
V. P. Kuznyak, Oleg K. Kolesnytskyj · Optoelectronic Information-Power Technologies · 2024
The article provides an analysis of known classes of methods for predicting side effects of polypharmacy. A new method of predicting the side effects of polypharmacy based on a heterogeneous graph neural network with blocks of attention is proposed. Based on known information about the drug, namely individual side effects and interaction with protein receptors, the network is able to predict the presence of side effects when combined with other known drugs. This information, in the form of a graphical representation of the data for each of the two drugs, is fed to the neural network, which determines the presence of a connection between the two nodes and the probability of each side effect given during training. The network, due to its inductive properties, is able to make predictions for drugs that were not used during model training, providing the ability to generalize side effect predictions for any drug with known individual side effects and target protein information.