Hi all, I have a tool that creates and fixes meshes. It can start from bad tracings and photographs. It works great for me, now I’d like to make sure it works for you too. If you’d like to link me to the worst of your source material, I’ll see if I can fix it and create a perfect mesh for you.
For context, I’m working on finite element models of astrocytes. “Intricate” meshes, 10 nm to the nearest neuron. I’ve been following the work of Jaeger & Tveito out of Norway, they’re right about one thing: charge matters.
But it’s come to my attention that meshing is needed in many areas of neuroscience and people are having problems with it, especially starting from photographs and tracings.
So, link me to your dogs, the worst of the worst, and let’s document the limits of what this tool can do, and improve it as needed. So far it’s doing an impressive job, it even works on half baked tracings from Blender. There are some things it just can’t do (GIGO, as they say), but it’ll fix holes and gaps in the tracing and it’ll attach disconnected segments and things like that. If you’re using meshparty you’ll want to run it through this tool “first”, and then pass the result to meshparty, you’ll be very happy with what comes out.
I’m still trying to wrap my mind around this. MICrONS started with EM and ended up with SWC files.
My first thought was “no one bothered to map the organelles?” Then upon studying the file formats I realized it was impossible! Which is a huge problem, because of stuff like this:
https://www.researchgate.net/figure/The-spatial-distribution-of-the-ER-dictates-Ca-microdomain-activity-in-perisynaptic_fig5_396526491
“Clustering matters”. The arrangement of ER in terminal processes varies by cell type. It looks different in inhibitory vs excitatory synapses. It’s a forehead-slapper, you’re going through all this work doing amazing automated analyses and this is what we end up with? Unusable SWC files with no organelles? Those of you doing biophysics know exactly what I’m talking about, it’s painful creating a usable computational mesh and it doesn’t have to be that way. That’s why a lot of us give up and end up using ball and stick models or NEURON compartments or something.
My SWC tool will get you from a skeleton to a mesh. If the goal is computational usability combined with realism, then it’s probably okay to have a missing segment, just delete it and use the other end of the cell. (You’re not going to use the whole cell anyway, maybe a branch of an astrocyte with a few PAP’s). However the tool will visualize the offending segment(s) and you can decide what to do (it’s equally risky to join or delete). Since SWC files seem to be the standard, the tool is capable of repairing the misalignments that typically occur near the soma, by brute force and using heuristics. Overall if you have 1% changes you’re probably safe, if you have 5% changes start over with a different cell.
I wonder if Allen would be amenable to a data mining experiment to extract the organelles from the original source material. Given coordinates, my tool will create an OBJ file that faithfully maintains the internal organelles as separate objects. (How good is the source? Can we distinguish ER cisternae and microvesicles in PAP’s? :eyebrow: ) You can import the file directly into Blender, Blender is good for at least 10k objects. As long as you export only from Blender (and not from Meshparty which doesn’t respect the OBJ spec) you can move seamlessly back and forth. Barring a new standard, this is as close as we can currently get to computations with organelles, unless you roll your own geometry.