Figure 2.

The steps of representation learning in our model. (a) Integrating different dimensions of biomedical database into a heterogeneous network and representation learning with their local content features. (b) The end-to-end transformer for chemical SMILES representation learning. (c) Aggregating the local content features to global features through the heterogeneous network using the graph neural network. CTD: Comparative Toxicogenomic Database; SMILES: Simplified Molecular-Input Line-Entry System.