From molecular structure to scalable self-assembly simulation

The question

How do we turn a molecular structure into a usable model—and simulate it at larger scales without losing numerical correctness?

Preparing a structure-resolved reaction-diffusion model requires identifying interfaces, defining coarse-grained geometry, and enumerating reactions. ioNERDSS connects those steps in a Python workflow.

Large particle-based simulations introduce an additional challenge: reactions and assemblies must remain consistent across processor boundaries. NERDSS-MPI addresses this through spatial domain decomposition and distributed communication.

The approach

1. 01

   Build the model

   Convert PDB/mmCIF structures into coarse-grained models using KD-tree interface detection, reaction generation, and sequence alignment.

2. 02

   Distribute the work

   Partition space across MPI processes and communicate particle, reaction, and assembly information across domain boundaries.

3. 03

   Verify the result

   Compare diffusion, reactions, assembly, dissociation, and restart behavior against analytical theory and serial simulations.

What I developed

• Implemented C++/MPI spatial domain decomposition and distributed communication.
• Validated distributed algorithms across seven benchmark models spanning diffusion, membrane reactions, and clathrin assembly.
• Developed automated structure processing and reaction-model construction, including ML-assisted affinity initialization.
• Released Python simulation, restart, analysis, and visualization workflows with tests, packaging, and documentation.

• C++
• Python
• MPI
• HPC

Results & validation

• Approximately 90× speedup on 96 CPUs for a 20,000-particle benchmark; this is a measured workload-specific result, not a universal scaling guarantee.
• Structure-processing workflows benchmarked on 44,000+ structures and assemblies ranging from 3 to 720 subunits.
• Distributed results were checked against analytical theory and serial simulation across seven benchmark models.

Related publications