About Me
I am a computational chemist with a passion for developing tools to accelerate scientific discovery. I specialize in machine learning interatomic potentials (MLIPs) and their application to ultra-large-scale molecular dynamics simulations of reactive chemical systems, earning me a PhD in Chemistry from the University of Florida in 2025.
It's hard to pinpoint when I got my start, but if I had to pick an exact moment, it would be when I was seven years old playing Sly Cooper on my PlayStation 2. This semi-open-world game had me exploring the boundary conditions of the game world and playing around with the physics the game modeled. That simple life experience led me down a path of loving video games and having a constant curiosity about the way physics are implemented in those virtual worlds.
Fast forward several years to my college years, I found myself studying chemistry and astronomy, where I learned that I could combine my love for understanding the natural world with programming and how we can use computational tools to solve massive challenges in the sciences. I joined Dr. Bill Miller's research lab at Truman State University, where I proposed my own project: Simulating selective water filtration using functionalized carbon nanotubes. This project introduced me to molecular dynamics, system design, and learning the ropes of high-performance computing and scripting.
This lead to my graduate studies at the University of Florida in the lab of Dr. Adrian Roitberg, where I developed frameworks for training and deploying machine learning interatomic potentials (MLIPs) to accelerate large-scale molecular dynamics simulations of complex chemical systems. In my first meeting with Dr. Roitberg, he showed me a simulation of carbon atoms assembling into graphene sheets, and I was hooked.
Since then, I've worked on a variety of projects, from simulating prebiotic chemistry to developing tools for uncertainty quantification in machine learning potentials and extending the ANI models to commonly used simulation packages. My work has been published in peer-reviewed journals, and I've been lucky to present my research at several conferences.
My biggest passion in life is learning; if you tell me something new and intriguing, you can bet that I'll open up a new tab as a starting point to research more deeply later. Outside of that, I enjoy hiking, comic books, playing video games, and experimenting with new recipes in the kitchen (especially ones that earn me compliments from my partner).
My Contact Information
Side Projects
LLMini
Miniature GPT-style model for learning transformer internals and a path toward chemical token models (using SMILES/SELFIES).
ani-mm
Implementation of ANI MLIPs into OpenMM for running molecular dynamics with a live-viewer of the simulation.
Mythic Depths
Procedurally generated dungeon crawler game built with PyGame.
C++ RNG Simulator
Tool for building C++ random number generators for implementing into Mythic Depths (as well as applications in scientific computing).
Research
- TorchANI: Atomic behavior in MLIPs, uncertainty quantification, extending functionality and integration into popular MD suites (LAMMPS, OpenMM).
- cuMolFind: Graph-based, GPU-accelerated molecular dynamics analysis toolkit.
- LUKE: Use the Forces: Uncertainty-based toolkit for sampling molecular substructures.
- Early Earth Hero Run: Analysis scripts and workflows for the 22.8-million atom Miller-Urey Experiment simulation.
Publications
- Nicholas S. Terrel, Jinze Xue, Ignacio J. Pickering, Melisa Alkan, Adrian E. Roitberg. Exploring Prebiotic Chemistry with ANI Neural Network Potential. Manuscript in progress.
- Jinze Xue, Nicholas S. Terrel, Ignacio J. Pickering, Adrian E. Roitberg. LAMMPS-ANI: Large Scale Molecular Dynamics Simulations with ANI Neural Network Potential. ChemRxiv (preprint). 2025. DOI: 10.26434/chemrxiv-2025-8v03m
- Ignacio J. Pickering, Jinze Xue, Kate Huddleston, Nicholas S. Terrel, Adrian E. Roitberg. TorchANI 2.0: An Extensible, High-Performance Library for Neural Network Potential Design. Journal of Chemical Information and Modeling. 2025, 65 (21) 11656-11671. DOI: 10.1021/acs.jcim.5c01853
- Mikayla Y. Darrows, Dimuthu Kodituwakku, Jinze Xue, Ignacio J. Pickering, Nicholas S. Terrel, Adrian E. Roitberg. LEGOLAS: a Machine Learning Method for Rapid and Accurate Predictions of Protein NMR Chemical Shifts. Journal of Chemical Theory and Computation. 2025, 21 (8) 4266–4275. DOI: 10.1021/acs.jctc.5c00026
Conference Presentations
-
LAMMPS-ANI Hero Run: Simulating Early Earth Chemistry at an Unprecedented Scale
-
Modeling of Early Earth Chemistry: 22.8 Million Atoms Simulated with TorchANI
-
Uncertainty-Driven Data Generation in ANI Neural Network Potentials: An Atomistic Force-Based Approach
-
Atomistic Uncertainty Estimation in ANAKIN-ME Neural Network Potentials
-
Atomistic Uncertainty Estimation in ANAKIN-ME Neural Network Potentials
Volunteering
-
Open-source contributor (Python libraries for ML in STEM)
Contributions, PRs and code reviews, and user support for scientific Python tooling and ML-for-chemistry projects (bug fixes, performance improvements, documentation, and reproducible examples). -
Recruitment Ambassador
Served as a recruitment ambassador for the University of Florida, Department of Chemistry to help prospective graduate students learn about research opportunities and see the benefits of joining the UF Chemistry community.
I also served as a recruitment ambassador at Truman State University, sending admissions acceptance packages, conducting campus tours, and hosting prospective students during open house events. -
Mentorship & peer support
I have a passion for mentoring and supporting early-career researchers. I have provided guidance on Python workflows for computational chemistry and HPC, including debugging, environment setup, and best practices for reproducible research.
There is something about making complex topics "click" for someone that is less-than-technical that I find deeply rewarding. Perhaps it is due to my own experiences as a self-taught programmer coming from a non-traditional background.