Our collaboration on a chemical language model is out !

Great collaborative work with IBM research and ETH Zürich is now published in Digital discovery. The model for molecular property prediction is validated on the large proprietary toxicity dataset from our previous study (Webel, 2020), uses uncertainty to improve reliability, reveals cytotoxic motifs via attention and outperforms existing approaches!

You can check the paper on Born, 2023 and the code of the model is available here.