Plotting mito geometries by PCA with Bokeh and STL Viewer
a work in progress
Low mag
Mid mag
Hi mag
3D block
Data
HEK cells processed for S3EM (Os, UA, LA). 96 serial sections cut with Diatome knives on a Leica ultramicrotome, collected on a silicon chip. Imaged on a Zeiss Sigma VP SEM using a Gatan BSD with ATLAS5 control system (FIBICS) with voxel size of 4-4-70 nm (x-y-z).
Ground truth segmented in VAST Lite. Deep learning and inference conducted using Gunpowder +. Meshes generated using marching cubes at scale 2. Meshes cleaned up using GAMer2/BlendGAMer and exported Blender. Geometry paramaterization in Python using numpy-stl library. PCA was computed using Python sklearn library from curvature, surface area, and volume parameters.
Interact
Interactive scatter plot of principal components is generated using Bokeh in Python to produce a browser-ready embedding. The STL-Viewer library is linked to the Bokeh HTML embedding, using the Bokeh Div element class targeted from JQuery, to collect selected object name and load it as input into a STL-Viewer scene.
Bokeh has tools for Zooming (box zoom and scroll zoom) and Tap Selecting enabled. Play with the tools to the right of the plot.
Tap Select a point on the PCA and click the Update Model button to see the corresponding mitochondrion. Explore the principal component axes: what best explains variance in these samples?
Click Load Big Model button to load all mitos into a single scene (may take some time).