Fig. 3
From: CardioGenAI: a machine learning-based framework for re-engineering drugs for reduced hERG liability
Illustration of the forward pass of the CardioGenAI discriminative models. The graph representation of a given SMILES string is encoded by A a graph attention network (GAT). The B transformer-derived and C fingerprint feature vectors are encoded by feed-forward networks. These three encodings are then concatenated and passed to D a final feed-forward network to generate a prediction