Abstract
Generative modelling has recently emerged as a promising tool to efficiently explore the vast chemical space. In homogeneous catalysis, Transition Metal Complexes (TMCs) are ubiquitous, and finding better TMC catalysts is critical to a number of technologically relevant reactions. Evaluating reaction rates requires expensive transition state (TS) structure search, making traditional library-based screening difficult. Inverse-design of TMCs with a model capable of generating good TS guesses can lead to breakthroughs in catalytic science. We present such generative model herein. The model is an instance of an equivariant conditional diffusion model, and the key innovation lies in its specific data representation and training procedure, that allow generic databases (e.g. non-TS structures) to be leveraged at training time, while offering the desired controllability at sampling time (e.g. ability to generate TSs on demand). We demonstrate that augmenting the training database with generic (but related) data enables a practical level of performance to be reached. In a case study, our model successfully explores the chemical space around Vaska’s complex, where the property of interest is the H2-activation barrier, in two distinct settings: generation from scratch, and redesign of a specific ligand in a known TMC. In both cases, we validate a selection of novel samples with Density Functional Theory (DFT) calculations.
Original language | English |
---|---|
Title of host publication | Proceedings of the 38th Conference on Neural Information Processing Systems (NeurIPS 2024) - AI4Mat Workshop |
Number of pages | 10 |
Publication status | Accepted/In press - 2025 |
Event | 38th Conference on Neural Information Processing Systems - Vancouver, Canada Duration: 10 Dec 2024 → 15 Dec 2024 |
Conference
Conference | 38th Conference on Neural Information Processing Systems |
---|---|
Country/Territory | Canada |
City | Vancouver |
Period | 10/12/2024 → 15/12/2024 |