Machine Learning Workflows for Chemistry: Applications in Catalysis and Ionic Liquids

Aug 2, 2026

Upload by Ezra

Rapid growth of scientific data and breakthroughs in computer science have transformed chemistry through predictive machine learning. This thesis traces the shift from statistical models to modern machine learning methods, focusing on selectivity prediction in organocatalysis and property prediction for ionic liquids. It introduces Pythia, an accessible toolkit using 2D/3D descriptors and shallow learners to evaluate catalysis selectivity. The work also examines machine-learning-based prediction of viscosity and solubility, supporting faster, more targeted experimental discovery and better extraction of chemical insights.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages234

Reading Time9 h 50 min

AccessFree to download

Abstract

Author’s Declaration

Acknowledgments

Data Availability

Publications

Machine Learning Workflows for Chemistry: Applications in Catalysis and Ionic Liquids Stamatia Zavitsanou Oriel College University of Oxford July 2023 Author’s Declaration The work presented in this thesis was conducted under the supervision of Professor Fernanda Duarte at the Department of Chemistry, University of Oxford. I declare that all the work is my own, unless otherwise stated, and has not been submitted for any other degree at this or anyother university. Stamatia Zavitsanou July 2023 Abstract In today’s world, data is being generated and accumulated at an astronomical rate, presenting new opportunities and challenges for the scientific community. In parallel, advancements in computer science have revolutionized the landscape of chemistry. The confluence of these two fields has given rise to a wave of sophisticated machine learning algorithms capable of building powerful predictive models. The field of computational chemistry is now finding itself navigating through this rapid evolution of technological progress. This Thesis traces the progression from statistical models to modern machine learning techniques and is setting the stage for the intricate dance between data science and chemistry. The focus narrows down to the specific utilization of machine learning for selectivity predictions in organocatalysis and property predictions for ionic liquids. It introduces Pythia, a machine learning toolkit designed with accessibility in mind, aiming to democratize the application of machine learning in computational chemistry. Pythia employs 2D and 3D descriptors and shallow learners to predict selectivity for organocatalytic reactions. The power of Pythia is put to the test and its potential for predicting selectivity in catalysis is explored. This demonstrates the toolkit's practical utility in facilitating more efficient and targeted experimentation in the search for effective catalysts. Finally, we delve into the prediction of viscosity and solubility in ionic liquids, further highlighting the capabilities of machine learning in streamlining the prediction of chemical properties. This Thesis promises to accelerate the pace of discovery in computational chemistry, allowing scientists to handle the influx of data more efficiently and extract meaningful insights from it. Acknowledgments I would like to express my gratitude to Fernanda Duarte, for her support, supervision, and guidance throughout these past years. I am truly grateful to her for believing in me and granting me the opportunity to pursue my studies at the University of Oxford. I am sincerely thankful to the Department of Chemistry for funding my DPhil studies, and I would like to acknowledge the generous support from the Holly Synod of Greece, which funded the final year of my DPhil. Their financial assistance has been crucial in enabling me to pursue my research endeavors. I would like to extend my special thanks to Tom Watts and Emanuele Casali for the close collaboration on this Thesis. Their dedicated partnership and contributions have been instrumental in the development of this research. Additionally, I express my deep appreciation to Alistair Sterling (loukoumaki), Tom Young (zouzounaki), and Tanya, for being with me since the beginning. Thank you to Ally, Bernie, Tomasz, Hafiz, Veronika, Henry, Aleksy, Tristan, Hanwen, Chloe, Ewa and all other members of the group, their invaluable advice and guidance have been indispensable to the progress of this work. I am also grateful to Dr. James McDonagh and Dr. Flaviu Cipcigian for their mentoring and support. Furthermore, I would like to acknowledge all my previous supervisors and teachers, particularly Dr. Zoe Cournia and Dr. Stavros Perantonis, for their significant role in motivating me to pursue this PhD. Lastly, I would like to express my deepest appreciation to my parents, Joanna and Nick, and my siblings, Rania and Iandros, as well as my theies and theios, Andreas, Diamado, Toula

FAQs

What are the main chemistry problems addressed in this thesis?

The thesis targets selectivity predictions in organocatalysis and property predictions for ionic liquids, including viscosity and solubility.

What is Pythia and what does it do?

Pythia is a machine learning toolkit designed to be accessible in computational chemistry. It uses 2D and 3D descriptors with shallow learners to predict organocatalytic selectivity.

How does the thesis evaluate the practical value of the proposed approach?

It tests Pythia’s power for predicting selectivity in catalysis and relates the results to more efficient, targeted experimentation for finding effective catalysts.

Aug 2, 2026

Upload by Ezra