Fig. 2: Accurate detection of purified 70S ribosomes from densely populated tomograms.
From: MiLoPYP: self-supervised molecular pattern mining and particle localization in situ
We analyzed tilt series from the EMPIAR-10304 dataset using our content exploration and particle localization modules. a, 2D grid visualization of the embedding space (top, left) showing ribosomes (red), gold fiducials (purple), as well as other patches corresponding to contamination and background areas. A selection of patches containing gold fiducials (purple outline) and 70S ribosomes (red outline) is shown below. 3D tomogram visualization view (top, right) with voxels colored according to the embedding coordinates shown in the 2D grid visualization representation. The exploration module is able to successfully differentiate gold fiducials from ribosomes despite their similar size and shape. b, A partially annotated input tomogram showing 70S ribosomes (blue circles) used for training the refinement model (left). The trained model accurately identifies ribosomes (red circles) while ignoring gold beads and contamination/background areas (right). c, 5.0 Å resolution reconstruction of the 70S ribosome obtained from a total of 6,051 particles using SPT.
