Speaker
Description
Chemical compound space refers to the vast set of all possible chemical compounds, estimated to contain 1060 molecules. While intractable as a whole, modern machine learning (ML) is increasingly capable of accurately predicting molecular properties in important subsets. This has had a massive impact on molecular design, since ML models can be used to screen unprecedentedly large databases. However, chemistry is at its core a science of transformations. Unfortunately, the space of possible chemical reactions is even more daunting, making a comprehensive understanding of reaction space highly challenging. This makes the use of ML in this context an even more interesting prospect. In the past years, we have therefore extensively engaged in the data-driven study of chemical reaction space, following a two-pronged approach.
First, this concerns the algorithmic enumeration of elementary reactions in a chemical subspace [1]. Here, a graph-theoretical approach was developed, which allows iteratively deriving increasingly complex reactions for a given set of intermediates. In this manner, unbiased reaction networks can be generated, which may span millions of steps. In most cases, such complex networks are likely over-complete, however, containing many irrelevant reactions for a given process. They thus merely form the basis for a subsequent reduction to a core network of the most important steps.
To achieve this, we have therefore developed tailored ML models for predicting reaction energies [2]. As an important basis for ‘reactive’ ML, we established a first-principles database (Rad-6) containing closed and open-shell organic molecules, along with an associated database of chemical reaction energies (Rad-6-RE). We showed that the special topology of reaction spaces, with central hub molecules involved in multiple reactions, requires a modification of existing “compound space” ML-concepts. Showcased by the application to methane combustion, we demonstrate that the learned reaction energies offer a non-empirical route to rationally extract reduced reaction networks from large reaction spaces.
[1] J.T. Margraf and K. Reuter, ACS Omega 4, 3370 (2019).
[2] S. Stocker, G. Csányi, K. Reuter, and J.T. Margraf, Nat. Commun. 11, 5505 (2020).
| Abstract Number (department-wise) | TH 01 |
|---|---|
| Department | TH (Reuter) |
