Parse structures
Read PDB or mmCIF complexes, identify subunits and interaction interfaces, and regularize repeated copies.
All projects
Scientific infrastructure
From atomic structures to scalable self-assembly simulation
Structure processing, coarse-grained model construction, and MPI reaction-diffusion
44k+
structures evaluated
96
CPU scaling target
2D + 3D
reaction-diffusion simulations
Context
The problem
Structure-resolved reaction-diffusion models require substantial manual work before simulation: identifying repeated subunits and interfaces, defining rigid-body geometry, enumerating reactions, assigning rates, and verifying that the resulting model is physically consistent.
Large particle-based systems add a second problem: explicit diffusion, reactions, and higher-order assemblies are computationally expensive and must remain accurate across processor boundaries.
Method
How the system works
Build models
Construct coarse-grained rigid bodies, reaction definitions, baseline kinetics, and optional affinity estimates.
Run simulations
Execute particle-based reaction-diffusion in 2D or 3D with explicit association, dissociation, and diffusion.
Scale with MPI
Partition space across processes, communicate ghost-region data, and synchronize reactions and assemblies across domains.
Role
My contribution
- Designed and co-developed ioNERDSS as a reusable Python structure-to-simulation workflow.
- Integrated interface detection, repeated-subunit regularization, reaction enumeration, deterministic ODE baselines, and ML-assisted affinity initialization.
- Designed and implemented MPI spatial domain decomposition and ghost-region communication in the C++ NERDSS engine.
- Validated parallel behavior across reversible reactions and higher-order self-assembly in both 2D and 3D.
Result
Outcomes and scientific value
✓
Validated structure-processing workflows across more than 44,000 PDB structures.
✓
Achieved close-to-linear scaling to 96 CPUs for suitable simulation regimes while preserving reaction-diffusion accuracy.
✓
Connected model construction, simulation, and analysis in a workflow usable by both developers and domain scientists.
Resources
Papers, code, and documentation
Related work
Publications
2026
Preprint · bioRxiv · 2026.01.27.702082
Transforming Macromolecular Structures into Simulations of Self-Assembly
Contribution: Structure-processing workflows, software development, and validation.
2025
Journal article · Journal of Computational Chemistry
Parallelization of Particle-Based Reaction–Diffusion Simulations Using MPI
Contribution: Lead software development, parallel algorithm design, validation, benchmarking, analysis, and writing.