Our paper on active learning for discovering novel metal organic frameworks for carbon capture was award “Best Paper” at this year’s IEEE International Conference on e-Science.

This paper builds on prior work where we designed a large-scale workflow that uses GenAI to create new molecular structures. This workflow incorporates chemistry calculations from various tools (e.g., LAMMPS, CP2K, RASPA) to assess the quality and chemical feasibility of each candidate structure to filter out unrealistic or unattractive structures. This workflow was designed to enable GenAI to search some latent chemical space with minimal restrictions.

The results of this prior work proved promising and that novel discovery can be done with GenAI. However, this recent paper posits that active learning (AL) methods can be used to improve the efficiency of the aforementioned workflow. Our results in this paper showed the practical effectiveness of this idea: on average, the non-AL version of the workflow would generate 281 high-performing structures whereas the AL version would generate 604.

A full preprint of the paper can be found here.

Picture of Dr. Nathaniel Hudson receiving the best paper award at the 2025 e-Science conference.