CryoLens

CryoLens is a 3D variational autoencoder for generative reconstruction of cellular structures from cryo-electron tomography data (available on the CZI Virtual Cell Platform).

The basic idea is to learn a compressed representation of 3D subtomograms that captures structural variation, which you can then use for things like visualizing structural representations.

We’ve trained a generative model on a set of uniform structures on a large synthetic dataset composed of 100+ diverse structures. The model has been trained at scale and can produce high fidelity single-angle reconstructions. Tools are available for napari and the Virtual Cell Platform.