Augmenting Large Language Models for Automated Discovery of F-Element Extractants.

Area of Science:

• Nuclear Chemistry
• Computational Chemistry
• Materials Science

Background:

• Efficient separation of f-elements (lanthanides and actinides) is crucial for advanced technologies but hindered by their chemical similarity.
• Developing selective solvent extraction reagents for these elements is a slow and challenging process.

Purpose of the Study:

• To present an AI-enabled workflow for the rapid design and computational screening of selective extractant ligands for f-elements.
• To accelerate the discovery of novel ligands with enhanced selectivity for challenging separations like Am(III)/Eu(III).

Main Methods:

• Utilized a large language model (SAFE-MolGen) for guided molecular design and preliminary ligand ranking.
• Employed a supervised machine learning model trained on experimental data to predict ligand performance under realistic conditions.
• Integrated a pipeline for constructing 3D metal-ligand complexes and performing quantum mechanical free energy calculations to assess selectivity.

Main Results:

• Demonstrated the workflow for Am(III)/Eu(III) separations, a critical challenge in nuclear waste management.
• Identified several newly designed ligands predicted to offer higher Am(III)/Eu(III) selectivity compared to the benchmark extractant CyMe4BTBP.
• Successfully accelerated the computational exploration of molecular space in this data-sparse field.

Conclusions:

• The AI-enabled workflow significantly speeds up the generation and evaluation of novel lanthanide and actinide extractants.
• This approach provides a general strategy for designing selective ligands, overcoming limitations in traditional methods.
• The developed ligands show promise for improving the efficiency of f-element separations in various technological applications.