Projects

Selected software projects. Full list on GitHub.


CheMeleon-Rxn overview

CheMeleon-Rxn

Pre-trained reaction graph neural network foundation model for low-data reaction property prediction. Adapts descriptor-regression pre-training to condensed graphs of reaction, achieving best-or-tied-best performance on seven of eight benchmark tasks spanning activation energies, enthalpies, yields, and rate coefficients.

Python PyTorch Graph Neural Networks Cheminformatics

OpenADMET-ExpansionRx

Multi-architecture benchmarking pipeline for predicting drug absorption, distribution, metabolism, and toxicity (ADMET). Spans gradient-boosted trees, message-passing neural networks, and foundation models. Placed 25th of 100 finalists in the OpenADMET-ExpansionRx blind challenge across 9 endpoints. Uses clustering-based scaffold splits, temporal splits, and statistical hypothesis testing for architecture selection.

Python Cheminformatics RDKit Drug Discovery

Polyelectrolyte simulation snapshot

DDPM-Enhanced-Sampling

Generative framework using denoising diffusion probabilistic models (DDPMs) to sample polymer conformations consistent with Boltzmann statistics. Replaces expensive enhanced-sampling simulations with a learned model that generates physically valid conformations orders of magnitude faster.

Python PyTorch Diffusion Models MD Simulation

Collinear swimmer in potential flow

Swimming-in-Potential-Flow

Companion code for the Journal of Fluid Mechanics paper on self-propulsion in potential flow. Implements boundary integral methods and analytical swimming theory in C++/CUDA for studying collective hydrodynamic interactions, demonstrating that viscous dissipation is not required for self-propulsion.

C++ CUDA Fluid Dynamics HPC

Polyelectrolyte surface adsorption simulation

Analysis-Polyelectrolyte-Surface-Adsorption

High-performance analysis pipeline for polyelectrolyte molecular dynamics trajectories. Custom multi-threaded MDAnalysis extensions achieve 100x speedup over serial processing for enhanced-sampling simulations. Automates free-energy surface construction from biased trajectory data.

Python MDAnalysis GROMACS HPC

Polyelectrolyte-Surface-Adsorption

Production MD simulation workflows for studying polyelectrolyte adsorption to mineral surfaces. Uses GROMACS and metadynamics to compare direct and water-mediated binding modes on CaCO₃, with applications to water treatment and scale inhibition.

Python GROMACS Molecular Dynamics Metadynamics