Jingyi (Victor) Li

ChironRNA: RNA structure refinement via E(3)-equivariant diffusionFirst author

• Designed and implemented a DDPM-based diffusion framework with E(3)-equivariant Graph Neural Networks (EGNN) in Python/PyTorch for resolving steric clashes and reconstructing missing atoms in RNA structures.
• Developed a hierarchical architecture combining all-atom and coarse-grained (five-point nucleotide representation) diffusion models, where the coarse-grained model recovers 23.7% of cases where the all-atom pipeline stalls.
• Implemented a conditional generation pipeline using MolProbity diagnostics to identify clash regions and selectively regenerate distorted substructures while preserving valid regions as constraints.
• Achieved 80% clash reduction on over 80% of the test set and a 100% atom reconstruction rate for structures under 200 nucleotides. Benchmarked on 90 RNA motifs with low RMSD and chemically accurate coordinates.

ATLAS: Large-scale 3D RNA motif library with graph searchFirst author

• Built an automated data pipeline to process 8,791 RNA structures from PDB, extracting and classifying 433,996 motifs across five categories (hairpin loops, bulges, internal loops, junctions, pseudoknots).
• Engineered three representations per motif: 3D atomic coordinates, nucleotide-level graphs with non-Watson-Crick interactions, and graphs without non-WC interactions. Implemented a graph isomorphism search algorithm for querying motifs by user-defined substructures.
• Revealed that non-WC interactions significantly increase structural diversity: tetraloop hairpins alone exhibit 1,487 unique conformations, most of which were invisible to prior libraries.
• Developed a web interface for community access to search, visualize, and download RNA motifs, supporting both canonical and non-canonical interaction patterns.

Molecular dynamics of plant epidermal cell wall mechanics

• Performed large-scale MD simulations in LAMMPS to investigate the mechanical behavior of cellulose microfibers in plant epidermal cell walls, building on the methodology of Zhang et al. (Science, 2021).
• Modeled shear stress-sliding relationships in cellulose bundles, capturing the telescopic sliding mechanism that enables cell walls to maintain both strength and extensibility during plant growth.
• Developed and validated force field parameters and simulation protocols for cellulose-hemicellulose systems, analyzing nanoscale mechanical responses under tensile and shear loading.

Diffusion-based molecular design in flexible protein pocketsCollaborator

• Contributed to methodology discussions for YuelDesign, an SE(3)-equivariant diffusion framework for structure-based drug design that models protein pockets as flexible, dynamic structures rather than rigid snapshots, including how pocket flexibility is represented in three-dimensional molecular generation. (Wang et al., Science Advances, 2026)

GNN-based protein-ligand binding site predictionCollaborator

• Contributed to methodology discussions for YuelPocket, a Graph Neural Network for unified protein-small molecule binding site prediction, including its formulation and its use of the large-scale PLINDER dataset for training. (Wang & Dokholyan, PNAS, 2026)

Data-driven constitutive modeling for brain tissue mechanics

• Developed Artificial Neural Network constitutive models in Python/MATLAB to predict the anisotropic mechanical behavior of brain tissue under intravascular pressure, building on the data-driven modeling framework of Tang, Z. et al. (Defence Technology, 2023).
• Integrated the ANN models into finite element simulations in Abaqus, achieving 95% accuracy while reducing computational cost by 80% compared with traditional hyperelastic models.

Nanoscale water transport on defective graphene surfaces

• Investigated the diffusion dynamics of water nanodroplets on double-vacancy graphene using MD simulations in LAMMPS, analyzing the constraining effects of defects on water mobility and wetting behavior. Published in Applied Surface Science (2021).

• Teaching assistant for EMCH 211 (Statics) and EMCH 212 (Dynamics) over two academic years, leading recitation sessions, grading assignments, and holding office hours.