Multi-modal Molecule Structure-text Model for Text-based Editing and Retrieval
Nature Machine Intelligence 2023

  • 1Mila
  • 2Université de Montréal
  • 3Nvidia Research
  • 4University of Illinois Urbana-Champaign
  • 5California Institute of Technology

  • 6Princeton University
  • 7HEC Montréal
  • 8CIFAR AI Chair
  • 9Arizona State University

  • Work done during intern at Nvidia Research

  • Equal advising
Text-based Molecule Editing for Binding.
  • Binding task from ChEMBL (ChEMBL 1613777). The text prompt is "This molecule is tested positive in an assay that are inhibitors and substrates of an enzyme protein. It uses molecular oxygen inserting one oxygen atom into a substrate, and reducing the second into a water molecule."
  • The visualizations are for the two output molecules edited with MoleculeSTM.
  • For the detailed structure of the input molecules, please check Figure 6.

MoleculeSTM pipeline

Downstream: Zero-shot Structure-text Retrieval

Downstream: Zero-shot Text-based Molecule Editing

Downstream: Molecular Property Prediction