We are fortunate to receive external support from the Wellcome Foundation via the Wellcome Leap Program, the United States Department of Energy, the National Science Foundation, the Research Corporation, and Google Research.

Nonequilibrium Self-assembly and Control

• Our group focuses on self-assembly in systems far from equilibrium, establishing fundamental physical limits on external control and demonstrating trade-offs between speed, accuracy, and energy consumption.
• We explore how external control and active driving forces can expand material design possibilities beyond equilibrium, providing theoretical frameworks and practical strategies for engineering self-assembled materials.

Key Publications

• Universal Energy-Speed-Accuracy Trade-Offs in Driven Nonequilibrium Systems
• Adaptive Nonequilibrium Design of Actin-Based Metamaterials: Fundamental and Practical Limits of Control
• Probing the Theoretical and Computational Limits of Dissipative Design

Funding: DOE

Biophysical Dynamics in and out of Equilibrium

• We have developed new computational methods using machine learning to predict transitions between molecular states, combining neural networks with smart sampling strategies for efficient learning from short simulations.
• We are currently working on reactive molecular dynamics for ATP hydrolysis using quantum computers, providing a framework for studying biological dynamics with significantly reduced computational requirements.

Key Publications

• Learning Nonequilibrium Control Forces to Characterize Dynamical Phase Transitions
• Microscopic Origin of Tunable Assembly Forces in Chiral Active Environments

Funding: Wellcome Leap, DOE

Statistical Inference and Sampling with Generative Models

• We have developed approaches for efficient molecular sampling using generative machine learning models, including coarse-graining with graph neural networks and transformer models for protein conformations.
• We built on our adaptive Monte Carlo approach using normalizing flows to propose nonlocal moves, significantly improving sampling efficiency and enabling the study of larger molecular systems and longer timescales.

Key Publications

• Data-Efficient Generation of Protein Conformational Ensembles with Backbone-to-Side-Chain Transformers
• Ensuring Thermodynamic Consistency with Invertible Coarse-Graining

Funding: Department of Energy, Google Research

Large-scale Machine Learning for Chemical Discovery

• We have developed machine learning approaches for molecular design, structure determination, and reaction mechanisms, including the Energy Rank Alignment (ERA) framework for efficient molecule generation
• We have developed multitask neural networks for predicting molecular structures directly from NMR spectra.

Key Publications

• Energy Rank Alignment: Using Preference Optimization to Search Chemical Space at Scale
• Accurate and Efficient Structure Elucidation from Routine One-Dimensional NMR Spectra Using Multitask Machine Learning

Funding: ResCorp