Fig. 1: Metabolic syndrome and application of a chemical language model for multi-target de novo design.
From: Automated design of multi-target ligands by generative deep learning
a The metabolic syndrome (MetS) is a multifactorial disease. Therapeutic effects against various aspects of the MetS can be achieved by nuclear receptor activation, enzyme inhibition and G-protein-coupled receptor (GPCR) modulation strategies. Designed polypharmacology addressing multiple mechanisms may provide synergies. b Chemical language models (CLM) are trained on molecules in string representation such as SMILES and can design new molecules in a data-driven fashion. Pretraining with a large set of molecules (e.g., from ChEMBL) allows the model to capture the syntax of SMILES. Fine-tuning with small sets of molecules can then be performed by transfer learning to bias the CLM toward designing molecules of interest. Here, we applied CLM to multi-target design by using sets of known ligands of six target pairs of interest for fine-tuning. a, b created with BioRender.com, released under a Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International license.
