Turning the dials on our new Hydra pFIB for the first time today. Can't wait to get some cells into it, soon! Been freezing things down like mad for the past couple of weeks.

It's alive! Still a ways to go but both beams are on.

This delivery marks the dawn of a new era for cryo-ET at the Penn State College of Medicine. What should we mill into first?

By dual training with synthetic data and very small amounts of real data, accurate and robust multi-class networks can be trained to segment intracellular components.

With CTS, it takes only minutes to produce annotated simulations for training models to segment real tomograms of purified macromolecular complexes.

By training regression networks on CTS-derived synthetic "cytoplasm" we generated a network that can #deepclean a wide range of real tomograms to a nearly noiseless state.

We’re excited to share cryo-TomoSim (aka CTS), our #cryoET modeling and simulation software, in addition to our new preprint documenting its use in training regressive denoising and semantic segmentation U-Nets. biorxiv.org/content/10.1101/…

Funny you should say that, because our segmentation models are beginning to make our tomograms look like early-stage @dsgoodsell paintings. Eventually I hope we can implement a "Goodsell" display-style within Dragonfly. Just throwing it out there🤔 @OrsDragonfly3D

Finally, this is how the same network performed on a neuronal tomogram from a different microscope, with a different camera and Volta Phase Plates. Membranes were manually colored to highlight the diversity of cell structures observed. EMDB Entry: EMD-10766

This segmentation of a pioneering microtubule was generated by the same network, and was collected during a different session on the same scope.

Networks are robustly inferable to data collected under similar conditions, such as this tomogram from the same data collection session. Membrane (cyan), F-actin (green), cofilactin (red), MTs (blue), ribos (orange) and TriC (pink) were all segmented by a single network.