We also compare Nellie to existing state of the art methods for both segmentation and tracking across imaging resolution, structure widths, and structure lengths, as well as the pipeline's speed on different sized images

🧵14/N

Now we can generate flow fields and get cool metrics, like if fields are diverging from (fission?) or converging to (fusion?) a point, or whether a branch is rotating along its long axis (?? dancing?). All of this, and much much more, linked to our organellar hierarchy.

🧵13/N

From these beacons, we can use a distance and cost-weighted interpolation to generate tracks for any coordinate within our organelles! Even at a sub-voxel level. We can use this to link voxels (and thus objects) between temporal frames too.

🧵12/N

We generate internal motion capture markers (think Hollywood) and pattern match these markers between frames to create beacons that point the way for the actual tracking. This means that we don't have to rely on finicky segmentations or skeletons for tracking.

🧵11/N

Then Nellie breaks down the organelles into connected components, and uses its skeleton to break it down even further into branches, which break down into individual nodes, which are variable-range containers for surrounding voxels. A big hierarchical organelle tree.

🧵10/N

So how does Nellie actually work? First Nellie figures out your image's metadata. Then it applies an automated structural-enhancement regime (modified Frangi filter + others) to make those organelles pop.

🧵9/N

And just for fun, here's me walking around, masked by Meta's Segment Anything Model and tracked via Nellie's tracking pipeline

🧵7/N

We used Nellie to unmix multiple organelles from a single fluorescence channel with high accuracy using just their morphology and motility patterns with a standard (random forest) classifier.
More fluorescence channels for your experiments!

🧵6/N

We also analyzed the intracellular ER highways using Nellie's outputs, allowing us to understand organellar rearrangements using its topology from the perspective of graph theory

🧵5/N

Inspired by Google Deepmind's weather forecasting model GraphCast, we also developed an organelle graph autoencoder using Nellie's outputs to analyze mitochondrial networks' responses to ionomycin treatment, revealing complex dynamics invisible to traditional analysis

🧵4/N

Nellie is FULLY automated with NO parameters to tune, and works on ALL structures - from mitochondria to ER to microtubules and beyond

(credit to @andrewgyork.bsky.social and @amsikking.bsky.social 's Snouty SOLS microscope for the data, and @alisterburt.bsky.social for the Animation plugin)

🧵3/N

Do you LOVE organelles??
Well I am THRILLED to announce that Nellie has been published in @naturemethods.bsky.social!

Nellie a fully automated pipeline for organelle segmentation, tracking, and hierarchical feature extraction in 2D, 3D, timelapse, multichannel live-cell microscopy

🧵1/N