Machine learning scientist Available now
Machine learning for scientific discovery.
I build learning systems that reason over geometry, structure, and physical constraints. My work connects geometric deep learning with structural biology, molecular simulation, and mechanics.
Jingyi Li, Ph.D. in Computational BiophysicsPenn State, 2026
01 / SELECTED RESEARCH
Evidence in the work.
Two first-author systems anchor my research. Both begin with a scientific representation problem and end with a usable, measurable result.
Geometric generative modeling
ChironRNA
Repairing RNA structures as a conditional generation problem.
Designed and implemented a hierarchical diffusion system with E(3)-equivariant graph neural networks. MolProbity diagnostics identify damaged regions, and the model regenerates those regions while preserving valid geometry.
- 80% clash reduction on more than 80% of the test set
- All-atom and five-point coarse-grained diffusion
- Python, PyTorch, EGNN, DDPM, MolProbity
Structural data and graph search
ATLAS
Turning the RNA structure archive into a searchable graph library.
Built an automated pipeline that converts experimental RNA structures into atomic coordinates and graph representations, then makes their substructures searchable through graph isomorphism.
- 8,791 PDB structures processed
- 433,996 motifs across five structural categories
- 1,487 distinct tetraloop hairpin conformations revealed
02 / RANGE
One method, more than one domain.
My foundation is mechanics and simulation. That background shapes how I build machine learning systems for scientific questions beyond a single molecule type or application area.
Represent the science
Geometry before architecture
Encode coordinates, interactions, symmetries, and constraints before choosing the learning system. My work spans molecular graphs, atomic structure, and continuum mechanics.
Build the system
Models plus the full pipeline
Develop data processing, training, conditional generation, diagnostics, benchmarking, and interfaces. I work primarily in Python and PyTorch on Linux and HPC systems.
Check the result
Scientific validation matters
Evaluate models using geometric quality, physical behavior, simulation, and domain baselines. A plausible output is only useful when its scientific claims survive inspection.
EARLIER SCIENTIFIC WORK
From mechanics to molecular learning
- Plant cell wall mechanicsLarge-scale LAMMPS simulations, force field development, and cellulose–hemicellulose mechanics.
- Data-driven brain tissue mechanicsNeural constitutive models integrated into Abaqus finite element simulations, reaching 95% accuracy with 80% lower computational cost.
- Nanoscale water transportMolecular dynamics of water nanodroplets on defective graphene, published in Applied Surface Science.
03 / BACKGROUND
Built across disciplines.
Training in mechanics, simulation, biophysics, and geometric learning gives me a practical vocabulary for collaborating across scientific and engineering teams.
Penn State University
Ph.D., Engineering Science and Mechanics
Computational biophysics · Graduate Research AssistantDalian University of Technology
MPhil, Engineering Mechanics
Data-driven mechanics · Molecular simulationDalian University of Technology
Bachelor, Engineering Mechanics
Mechanics and computational modelingResearch toolkit
- Learning
- Diffusion models, EGNN, GCN, CNN, conditional generation
- Code
- Python, PyTorch, JAX, NumPy, SciPy, pandas, C++, SQL
- Simulation
- LAMMPS, GROMACS, Abaqus, molecular dynamics, finite elements
- Structure
- PDB pipelines, RNA analysis, MolProbity, PyMOL, graph search
- Systems
- Linux, HPC clusters, SLURM, Bash, Git
04 / PUBLICATIONS
Selected publications.
First-author work is separated from collaborations so the role behind each result remains clear.
2026
First author
ChironRNA: Steric Clashes Resolution in RNA Structures via E(3)-Equivariant Diffusion
Jingyi Li, Jian Wang, and Nikolay V. Dokholyan · bioRxiv
2026
First author
ATLAS: Graph-based 3D RNA Motif Library Incorporating non-Watson-Crick Interactions
Jingyi Li, Jian Wang, Srinivasan Ekambaram, and Nikolay V. Dokholyan · bioRxiv
2021
Co-author
Diffusion of water nanodroplets on graphene with double-vacancy
Lei Deng, Jingyi Li, Shan Tang, and Zhen Guo · Applied Surface Science 573, 151235
05 / CONTACT
Let's build the next scientific model.
I am exploring machine learning scientist and research scientist roles where models are built in close contact with scientific questions, data, and validation.
jingyili.mechanics@gmail.com