Practical applications of machine learning to problems in catalysis

Rinehart, Nicholas Ian

Permalink

https://hdl.handle.net/2142/121338 Copy

Description

Title

Practical applications of machine learning to problems in catalysis

Author(s)

Rinehart, Nicholas Ian

Issue Date

2023-07-13

Director of Research (if dissertation) or Advisor (if thesis)

Denmark, Scott E

Doctoral Committee Chair(s)

Denmark, Scott E

Committee Member(s)

• Hergenrother, Paul J
• Sarlah, David
• Girolami, Gregory S

Department of Study

Chemistry

Discipline

Chemistry

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Machine learning
• catalysis, data science

Abstract

This thesis covers work which can broadly be categorized into modeling enantioselectivity and modeling yield. The challenges associated with each are different, but the overarching lessons in this thesis were derived from both projects. Our laboratory had figured out how to represent chiral catalyst shape when I joined, but the work laid out here addresses the next problem: creating the right dataset to train models capable of generalizing to new reactants. Models must be trained using a dataset comprised of a broad range of substrates using the same underlying logic that our laboratory identified for representing diverse catalyst structures. The first chapter of this thesis briefly describes the context in which the projects described in the subsequent two chapters began. It draws heavily from the Accounts of Chemical Research article that I was an author on in 2020, which chronicles the development of our laboratory’s chemoinformatic workflow and subsequent work performed by my colleague and mentor Dr. Andrew F. Zahrt. The second chapter describes a pilot project that I conducted in the laboratory which took place during that follow-up work. At the time, we had not demonstrated if our laboratory’s new descriptors would work for transition metal catalyzed transformations, or more generally for transformations which are known to have multiple, substrate structure-dependent, selectivity-determining interactions. I selected asymmetric hydrogenation to validate the descriptors in a complex system using literature data, and chapter 2 describes the lessons learned from that project. The third chapter describes a project in which I saw the same inherent problem that I had observed in the asymmetric hydrogenation work: we needed a better way to think about dataset design. The lessons learned in chapter 3 are no doubt applicable to the work in chapter 2 and, more generally, to enantioselectivity modeling.

Graduation Semester

2023-08

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/121338

Copyright and License Information

Copyright 2023 Nicholas Rinehart

Owning Collections