Abstract
The design of novel catalysts is an active field of chemical research, crucial for approximately 90% of industrial chemical processes. More efficient catalysts have the potential to decrease energy consumption, increase reaction yields, and enable currently unfeasible reactions, particularly those relevant to the green energy transition, such as power-to-x processes and carbon capture. Historically, the discovery of novel catalysts has been driven by trial and error and empirical observations.
Since the advent of reliable computational chemistry tools, highthroughput virtual screening and optimization algorithms, such as genetic algorithms, have been used to explore defined chemical spaces for promising catalysts. Relevant chemical constraints, such as stability and synthesizability, can be considered through careful selection of these chemical spaces. However, this approach does not facilitate the de novo discovery of catalysts.
In this context, machine learning-based tools offer a promising avenue for discovering novel chemical motifs and molecules. The realworld impact of these models on the design of efficient catalysts remains to be seen, as there is often no attempt at computational or experimental verification.
In the first part of this thesis, we present a method for the de novo discovery of efficient catalysts, moving beyond predefined chemical spaces using a graph-based genetic algorithm approach. We explicitly incorporate relevant chemical constraints and verify the success of the optimization computationally. Furthermore, we synthesize the catalyst and experimentally confirm its superior performance. This work represents a significant advancement towards more effective and efficient de novo catalyst design and its real-world application. Furthermore, we extend the approach to handle transition metal-based catalysts and show that we can efficiently find promising catalysts.
The second part of this thesis introduces an automated, fast, and user-friendly workflow designed to predict the regioselectivity of catalyzed C−H activations with directing groups. This workflow leverages semi-empirical quantum mechanical calculations to provide accurate predictions efficiently. By automating complex computational tasks, this approach streamlines the process of determining regioselectivity, making it accessible to researchers without extensive expertise in computational chemistry.
Since the advent of reliable computational chemistry tools, highthroughput virtual screening and optimization algorithms, such as genetic algorithms, have been used to explore defined chemical spaces for promising catalysts. Relevant chemical constraints, such as stability and synthesizability, can be considered through careful selection of these chemical spaces. However, this approach does not facilitate the de novo discovery of catalysts.
In this context, machine learning-based tools offer a promising avenue for discovering novel chemical motifs and molecules. The realworld impact of these models on the design of efficient catalysts remains to be seen, as there is often no attempt at computational or experimental verification.
In the first part of this thesis, we present a method for the de novo discovery of efficient catalysts, moving beyond predefined chemical spaces using a graph-based genetic algorithm approach. We explicitly incorporate relevant chemical constraints and verify the success of the optimization computationally. Furthermore, we synthesize the catalyst and experimentally confirm its superior performance. This work represents a significant advancement towards more effective and efficient de novo catalyst design and its real-world application. Furthermore, we extend the approach to handle transition metal-based catalysts and show that we can efficiently find promising catalysts.
The second part of this thesis introduces an automated, fast, and user-friendly workflow designed to predict the regioselectivity of catalyzed C−H activations with directing groups. This workflow leverages semi-empirical quantum mechanical calculations to provide accurate predictions efficiently. By automating complex computational tasks, this approach streamlines the process of determining regioselectivity, making it accessible to researchers without extensive expertise in computational chemistry.
| Originalsprog | Engelsk |
|---|
| Forlag | Department of Chemistry, Faculty of Science, University of Copenhagen |
|---|---|
| Antal sider | 114 |
| Status | Udgivet - 2024 |