Presentation abstract
In this talk, I will give a brief overview of the research done in my group on the field of machine learning (ML) applied to transition metal complexes (TMCs).
I will start with our initial work on predicting the energy barrier of the elementary steps involved in homogeneous catalysis, followed by our interest on generalizing ML approaches with tmQM, an 86K dataset of TMCs built from CSD data and semi-empirical quantum calculations, which we thereafter expanded with graph representations based on DFT and NBO.
After this short summary, I will introduce our recent work on evolutionary learning, including the tmQMg-L ligand library and the PL-MOGA algorithm. The tmQMg-L library contains 30K diverse and synthesizable ligands with defined charges and metal coordination modes. The PL-MOGA is a genetic algorithm allowing for the multiobjective optimization of TMCs within selected regions of the Pareto front, with fine control over both aim and scope. Used together, these tools allowed for the exploration and exploitation of vast chemical spaces containing billions of TMCs.
If time allows, I will present our current efforts in coupling the PL-MOGA to other generative models based on deep graph learning.
Time and place of the conference
May 22, 2024 at Heidelberg Congress Center, Germany.
For more information
Refer to the website: https://ccsc2024.github.io/