Fig. 1
From: CardioGenAI: a machine learning-based framework for re-engineering drugs for reduced hERG liability
The CardioGenAI framework for re-engineering hERG-active compounds. An autoregressive transformer decoder pretrained on a large dataset of SMILES strings generates compounds conditioned on the scaffold and physicochemical properties of a given input compound, and the generated ensemble is filtered based on desired activity against hERG, NaV1.5 and CaV1.2 channels. Cosine similarity is calculated between a 209-dimensional descriptor vector of the input compound and that of every filtered generated compound to identify the refined candidates most chemically similar to the input compound