Machine learning in chemical reaction space

Abstract

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. Here, we therefore engage in the ML-driven study of even larger reaction space. Central to chemistry as a science of transformations, this space contains all possible chemical reactions. As an important basis for ‘reactive’ ML, we establish 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 show 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 for detailed microkinetic analyses.

A review on machine learning algorithms for the ionic liquid chemical space

“…recently published a very interesting study about the use of ML in chemical reaction networks, which shows that the prediction of new data points using ML methods is performed much faster than with DFT calculations, with equal accuracy. 76 Using COSMO-RS s-proles as data for ML methods, seems promising and has been implemented in various classical property regression models with very promising results, [77][78][79] but so far with only few implementations to ML algorithms. [80][81][82][83][84][85] classication and prediction.…”

Machine learning potentials for extended systems: a perspective

“…Hence, nowadays, active learning strategies, as used in the ML community for quite some time [74], have become a standard approach for the construction of MLPs [75][76][77][78][79]. This search can be complemented by selecting non-redundant structural features from a large number of structures [80][81][82]. In both approaches the identification of relevant structures is even possible before carrying out expensive electronic structure calculations, allowing to focus the available computing time on the most important information.…”

Big data and machine learning for materials science

“…In doing so, the number of possible material structures is estimated to be 10 100 , which is larger than the number of particles in the universe [ 29 ]. In this vastness, the discovery of new materials must face resource and time constraints [ 30 , 31 ]. Expensive experiments must be well planned, ideally targeting lead structures with a high potential of generating new materials with useful properties.…”