Bridging the Gap Between Theory and Experiment with Machine Learning Open Access

Abstract

Accurately linking theoretical predictions with experimental measurements remains a central challenge in computational chemistry. Quantum mechanics provides a rigorous foundation for describing molecular systems, yet its direct application to realistic chemical environments is often computationally prohibitive. Machine learning has emerged as a powerful complement, capable of improving efficiency and predictive performance; however, many models sacrifice interpretability and robustness, particularly in noisy or data-scarce experimental settings. Overcoming these limitations requires approaches that enhance predictive accuracy, interpretability, transferability, and accessibility. In this dissertation, I address these challenges from several complementary perspectives. First, Δ-machine learning is employed to correct systematic biases in implicit-solvent (time-dependent) density functional theory calculations, thereby reducing functional dependence and improving agreement with experimental redox potentials and absorption energies. Second, to overcome the scarcity of experimental data, I develop a domain-adversarial framework that combines abundant simulated spectra with limited experimental measurements to detect thermodynamic phase transitions from angle-resolved photoemission spectroscopy. Applied to cuprate superconductors, this framework achieves over 97% accuracy in identifying phase transitions from experimental spectra. Third, I develop ACES-GNN, a graph neural network trained with activity-cliff explanation supervision that simultaneously improves predictive performance and interpretability. Finally, I develop AutoSolvate-Agent, a large-language-model-driven autonomous system that automates simulation setup, error recovery, and interactive interpretation for advanced solution-phase quantum chemistry.

Table of Contents

Chapter 1. Introduction ................................................. 1

 1.1 Computational methods for molecular property prediction .......... 2

 1.2 Representation of molecular and material system ................. 11

 1.3 Interpretability of machine learning models ..................... 19

 1.4 Outline of the dissertation ..................................... 24

Chapter 2. Δ-Machine learning for quantum chemistry prediction of 

solution-phase molecular properties at the ground and excited states ... 28

 2.1 Introduction .................................................... 29

 2.2 Methods ......................................................... 31

 2.3 Results and discussion .......................................... 39

 2.4 Conclusions ..................................................... 54

 2.5 Acknowledgments ................................................. 55

Chapter 3. Detecting thermodynamic phase transition via explainable 

machine learning of photoemission spectroscopy ......................... 57

 3.1 Introduction .................................................... 58

 3.2 Methods ......................................................... 61

 3.3 Results and discussion .......................................... 67

 3.4 Conclusions ..................................................... 79

 3.5 Acknowledgments ................................................. 81

 3.6 Appendix ........................................................ 81

Chapter 4. ACES-GNN: Can graph neural network learn to explain activity 

cliffs? ............................................................... 90

 4.1 Introduction .................................................... 91

 4.2 Methods ......................................................... 95

 4.3 Results and discussion ......................................... 102

 4.4 Conclusions .................................................... 115

 4.5 Acknowledgments ................................................ 116

Chapter 5. AutoSolvate-Agent: an autonomous agent for GPU-accelerated 

solution-phase quantum chemistry ...................................... 118

 5.1 Introduction ................................................... 118

 5.2 Methods ........................................................ 122

 5.3 Results and discussion ......................................... 125

 5.4 Conclusions .................................................... 130

 5.5 Acknowledgments ................................................ 130

Chapter 6. Conclusions and perspectives ............................... 131

Bibliography .......................................................... 137

About this Dissertation

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Laney Graduate School
Department	Chemistry
Degree	Ph.D.
Submission	Dissertation
Language	English
Research Field	Computer Science Physics, Condensed Matter Chemistry, Physical
Keyword	Computational Chemistry Machine Learning
Committee Chair / Thesis Advisor	Liu, Fang, Emory University
Committee Members	Evangelista, Francesco, Emory University Ribeiro, Raphael, Emory University

Last modified

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Bridging the Gap Between Theory and Experiment with Machine Learning ()	2026-04-15 10:58:51 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions