Research
Our research focuses on developing new mathematical tools, numerical algorithms, and computer software to qualitatively and quantitatively predict the outcome of chemical phenomena using strategies from quantum chemistry and machine learning. For an up-to-date list of our lab’s publications and collaborations, please visit our Google Scholar page.
1. Conceptual Quantum Chemistry
Bridging the gap between quantum mechanical calculations and chemical understanding, we develop rigorous mathematical definitions of empirical chemical concepts within the framework of quantum chemistry. Specifically, we focus on development and assessment of fuzzy atoms-in-molecules methods, based on the variational Hirshfeld framenwork, and combine them with energy decomposition analysis to gain chemical insight into the nature of molecular interactions and bonding. For relevant publications, see:
- 2026: Reverse Engineering Atomic Densities for Charge Model 5 (CM5) to Compute Atomic Properties Beyond Charges
- 2026: Permanent electrostatic moments through the lens of atoms: assessing variational Hirshfeld methods
- 2026: The Scaled Hirshfeld Partitioning: Mathematical Development and Information-Theoretic Foundation
- 2024: Variational Hirshfeld partitioning: General framework and the additive variational Hirshfeld partitioning method
- 2022: Constrained iterative Hirshfeld charges: A variational approach
2. Quantum-Mechanical Derived Force-Field Parameters
Improving the accuracy of molecular dynamics simulations, we develop methods to derive force-field parameters directly from quantum mechanical calculations. Our work primarily focuses on nonbonded interactions, which are essential for accurately modeling chemical interactions, including protein-ligand binding and the thermophysical properties of organic liquids. We further integrate these approaches with machine learning techniques for transferable force-field parameterization. For relevant publications, see:
- 2025: Nonbonded Force Field Parameters Derived from Atoms-in-Molecules Methods Reproduce Interactions in Proteins from First-Principles
- 2023: The Energetic Origins of Pi–Pi Contacts in Proteins
3. Machine Learning (ML) Potentials
Developing transferable and scalable ML potentials, we design chemically-informed physics-based ML models for molecular systems by adapting state-of-the-art graph convolutional neural network algorithms. Our approach enriches training data with chemically relevant information to improve both its quality and diversity, so that the models “learn better chemistry”. At the same time, we integrate physics-based models with ML to leverage the complementary strengths of both approaches, so that the models “learn better physics”. These transferable ML potentials are integrated into molecular dynamics simulations to enable large-scale, long-timescale simulations of chemical processes with near–quantum mechanical accuracy. For relevant publications, see:
- 2022: NewtonNet: a Newtonian message passing network for deep learning of interatomic potentials and forces
- 2020: Learning to Make Chemical Predictions: The Interplay of Feature Representation, Data, and Machine Learning Methods
4. Data-Driven Design of Gold Nanoclusters (AuNCs) for Cancer Therapeutics
Enhancing the efficacy of cancer therapies lowers treatment duration, improves survival rate, and reduces cost. Our research integrates machine learning (ML) and quantum chemistry (QC) methods to accelerate the discovery of novel photo-radiosensitizers that enhance the cancer-killing power of existing radiation and light-based therapies while minimizing damage to healthy tissue. Atomically precise gold nanoclusters (AuNCs) have emerged as a promising class of sensitizers, due to their stability, biocompatibility, and tunable properties. However, identifying the optimal AuNCs is challenging due to the vast number of possible candidates, the time- and resource-intensive nature of experimentation, and the accuracy-efficiency trade-off of traditional computational methods, which makes them unsuitable for large-scale screening. We address this challenge by developing hybrid QC/ML predictive models to identify the most promising AuNCs and accelerate AuNC-based therapeutic discovery.
5. QC-Devs Software Development
Making the most recent computational chemistry methods accessible to the broader scientific community for research and education, we develop free and open-source software tools for computational sciences as part of QC-Devs Inernational Software Consorsium. We follow the best practices in sustainable software development, including modular design, comprehensive documentation, rigorous testing, and Git version control, to ensure that our tools are user-friendly and customizable, to facilate testing ideas and support reproducible research and open science. Our software development efforts are focused on post-processing quantum chemistry calculations to gain chemical insight, as implemented in ChemTools software package. For relevant publications, see:
- 2026: Selector: A General Python Library for Diverse Subset Selection
- 2024: CuGBasis: High-performance CUDA/Python library for efficient computation of quantum chemistry density-based descriptors for larger systems
- 2024: Grid: A Python library for molecular integration, interpolation, differentiation, and more
- 2024: GBasis: A Python library for evaluating functions, functionals, and integrals expressed with Gaussian basis functions
- 2022: ChemTools: Gain chemical insight from quantum chemistry calculations