AI algorithm and dataset for metal-binding selectivity predictions

composite image: Computer-aided design software that predicts selectivity using a machine learning model and then ranks them based on selectivity is now available on GitHub.
Computer-aided design software that predicts selectivity using a machine learning model and then ranks them based on selectivity is now available on GitHub.

CMI researchers from Ames National Laboratory conducted the research for this highlight.

Innovation 
Ligand design code and two curated datasets with AI-friendly format that quickly produce high-accuracy selectivity predictions for molecules that bind to metal ions in water.

Achievement

  • Enabled for the first time fast and accurate prediction of metal-ligand binding for over 50 metal ions in water. 
  • Completed two well-curated and tested datasets and updated a published RE dataset, enabling machine learning. 
  • Provided free open-source code to enable further improvements and accessibility to the scientific community. 

Significance and Impact
Prediction of metal-ion binding affinity of new ligands lays the foundation for automated design of entirely new extractant molecules for selective separations, avoiding years of trial-and-error experimentation.

Hub Target Addressed 
Increasing the Speed of Discovery and Integration. Predictive Models for “Real” Materials. Workforce development.