Multiscale modeling and machine learning for molecular systems
Tristan Bereau’s group at the Institute for Theoretical Physics, Heidelberg University. We develop simulation and machine-learning methods for soft-condensed matter and biomolecules, and use coarse-graining to make chemical compound space tractable. Current activities include
- Free energies from physics-informed generative models
- Coarse-graining to compress chemical space for high-throughput screening
- Molecular discovery by active learning and Bayesian optimization
- Reproducible and FAIR workflows for molecular simulation
A fully funded PhD position on generative machine learning for molecular thin films is currently open, reviewed on a rolling basis.
What’s presented here is the concerted effort of many fantastic students, postdocs, and collaborators I’ve had the chance to work with.
Recent work
- Data-efficient multidimensional free energy estimation via physics-informed score learning
- Mapping the Molecular Universe: Exploring Chemical Compound Space by Multiscale High-Throughput Screening and Machine Learning
- Split-Flows: Measure Transport and Information Loss Across Molecular Resolutions
- Fast Parametrization of Martini3 Models for Fragments and Small Molecules
- Fokker–Planck Score Learning: Efficient Free-Energy Estimation under Periodic Boundary Conditions