3D SMLM point-cloud network analysis pipeline

Stages

1. 1.
   Blink merging preprocessing(in_silico_filter)

   To remove strongly interacting high-degree points before downstream network analysis.

   Selection: Removal of high-degree strongly interacting points

2. 2.
   Network comparison and blob retention(selection)

   To retain a sub-network of geometric structures or blobs for downstream quantitative classification.

   Selection: Comparison of Cav1 network properties with randomly generated networks to retain a sub-network of geometric structures

3. 3.
   Feature extraction and machine-learning classification(functional_characterization)

   To quantitatively describe each retained blob for downstream clustering and structural class identification.

   Selection: Extraction of 28 quantitative features describing size, shape, topology, and network characteristics

4. 4.
   Unsupervised clustering of structural classes(hit_picking)

   To identify distinct scaffold and caveolae classes from quantitative blob features.

   Selection: Unsupervised clustering of quantitative blob features

5. 5.
   Modularity-based assembly inference(secondary_characterization)

   To infer how smaller scaffold units may assemble into larger scaffolds and caveolae coats.

   Selection: Multi-threshold modularity analysis of scaffold relationships

Steps

1. 1.
   Remove high-degree points by iterative blink mergingpreprocessing algorithm

   Reduce strongly interacting high-degree points in SMLM localization data.

   This preprocessing occurs before network comparison so downstream structural analysis is performed on a cleaned point set.

2. 2.
   Compare Cav1 network properties with random networks to retain geometric blobs

   Retain a sub-network of geometric structures for downstream classification.

   This filtering step narrows the candidate structures before feature extraction and clustering.

3. 3.
   Extract 28 quantitative features from retained blobsanalysis pipeline

   Quantitatively describe blob size, shape, topology, and network characteristics.

   Feature extraction is needed before unsupervised clustering can identify structural classes.

4. 4.
   Cluster blobs to identify scaffold and caveolae classes

   Identify distinct structural classes among quantified blobs.

   Clustering follows feature extraction because class identification depends on the quantitative blob descriptors.

5. 5.
   Apply multi-threshold modularity analysis to infer assembly relationships

   Infer how smaller scaffold units may combine into larger scaffolds and caveolae coats.

   Assembly inference is performed after structural classes are identified so relationships among scaffold classes can be interpreted.