Machine-learning interatomic potentials
Accurate, transferable models for energies, forces, charge transfer, and dielectric response across complex chemical systems.
Research interests
From atomic interactions to autonomous discovery
I develop data-driven methods that make molecular and materials simulation more accurate, efficient, and capable of exploring new chemical spaces.
Graph deep learning
Equivariant and multi-fidelity graph neural networks that learn useful representations of molecules and crystalline materials.
Autonomous materials discovery
Multi-agent systems and closed-loop workflows that plan and execute high-throughput computational experiments.
Atomistic simulation
Combining electronic-structure calculations and molecular dynamics to uncover mechanisms in materials and interfaces.
Academic journey
Experience & education
My work spans theoretical chemistry, materials science, and artificial intelligence, with a consistent focus on connecting physical insight to scalable computation.
Postdoctoral experience
Distinguished Postdoctoral Fellow
Vector Institute & University of Toronto
Developing multi-agent systems for autonomous materials workflows, generative models across molecular and materials domains, and next-generation interatomic potentials for dielectric response.
Mentors: Alán Aspuru-Guzik & Anatole von Lilienfeld
Schmidt AI in Science Postdoctoral Fellow
University of California, San Diego
Led the development of MatGL and advanced multi-fidelity and foundation machine-learning interatomic potentials for materials simulations.
Mentor: Shyue Ping Ong
Selected publications
Research output
9First or co-first author
3Equal-contribution papers
5Corresponding author
- Preprint2026
El Agente Potente: High-Throughput Agentic Atomistic Simulations
T. W. Ko, J. Bai, T. Swanick, Y. Kang, C. Choi, A. Q. Jiang, A. Yin, V. Bernales, A. Aspuru-Guzik
arXiv:2609.14840 (2026)
An agentic framework combining typed execution graphs and coding workflows for high-throughput atomistic simulations.
- Accepted2026
A Linear-Scaling, Charge-Aware Foundation Potential for Atomistic Simulations
T. W. Ko*, R. Liu, A. R. Mishra, Z. Yu, J. Qi, S. P. Ong*
Nature Computational Science · Accepted
Introducing charge-equilibrated TensorNet (QET), a linear-scaling, charge-aware foundation potential for reactive atomistic simulation.
- 2025
Materials Graph Library (MatGL), an Open-Source Graph Deep Learning Library for Materials Science and Chemistry
T. W. Ko*, B. Deng, M. Nassar, L. B.-Luque, R. Liu, J. Qi, A. C. Thakur, A. R. Mishra, E. Liu, G. Ceder, S. Miret, S. P. Ong*
npj Computational Materials 11, 253
- 2025
Data-Efficient Construction of High-Fidelity Graph Deep Learning Interatomic Potentials
T. W. Ko*, S. P. Ong*
npj Computational Materials 11, 65
- 2023
Recent Advances and Outstanding Challenges for Machine Learning Interatomic Potentials
T. W. Ko, S. P. Ong
Nature Computational Science 3, 998
- 2023
Accurate Fourth-Generation Machine Learning Potentials by Electrostatic Embedding
T. W. Ko*, J. A. Finkler, S. Goedecker, J. Behler*
Journal of Chemical Theory and Computation 19, 3567–3579
- 2022
Neural Network Potentials: A Concise Overview of Methods
E. Kocer†, T. W. Ko†, J. Behler*
Annual Review of Physical Chemistry 73, 163–186
- 2021
General-Purpose Machine Learning Potentials Capturing Nonlocal Charge Transfer
T. W. Ko†, J. A. Finkler†, S. Goedecker*, J. Behler*
Accounts of Chemical Research 54, 808–817
- 2021
A Fourth-Generation High-Dimensional Neural Network Potential with Accurate Electrostatics Including Non-Local Charge Transfer
T. W. Ko*†, J. A. Finkler*†, S. Goedecker, J. Behler
Nature Communications 12, 398
Let’s connect