Fig. 1: Depiction of the workflow of the work reported here.
From: De novo design of polymer electrolytes using GPT-based and diffusion-based generative models
Starting from HTP-MD dataset, we tokenize the SMILES codes which represent the monomers of polymer electrolytes. Both unconditional generation (label-free) and conditional generation (conductivity class label inserted in the sequence) are performed with GPT-based and diffusion-based generative models. The generated polymer set is evaluated with different metrics, and promising candidates with high conductivity from conditional generation are further evaluated with MD simulations.
