Exploring the Microscopic World
Our mission is to develop principled, generalizable, and scalable AI models that address the fundamental limitations of cryo-EM data analysis.
We believe the next generation of cryo-EM algorithms will be driven by foundation models, probabilistic inference, and physics-aware learning, enabling robust reasoning about heterogeneity, uncertainty, and structure.
General-purpose density models for representation and generation.
Diffusion and flow-based modeling in 3D density space.
Physics-aware learning for ill-posed reconstruction tasks.
Learning continuous conformational landscapes from cryo-EM data.
Joint reasoning across atomic models and densities.
Data, benchmarks, and open model ecosystems.
CryoFM2 extends CryoFM by applying it to real experimental cryo-EM data, incorporating physics-aware process into RELION refinement and showing fine-tuning ability for various post-processing specific models.
CryoFM is a flow-matching-based foundation model that learns a generative prior over biomolecular density maps.
CryoSTAR a structural regularization approach for conformational heterogeneity in cryo-EM. It has been validated on diverse experimental datasets.
