Abstract
Quantitative analysis of network-like biological molecular architectures, such as cytoskeletal networks, remains a fundamental challenge in super-resolution fluorescence imaging because single-molecule localization microscopy (SMLM) typically produces sparse and discontinuous localization patterns arising from stochastic labeling, incomplete probe occupancy, and structural deformation of biological networks. These factors obscure the underlying connectivity, periodicity, and symmetry, limiting the ability of existing analysis methods to resolve higher-order organization. Here, we report the Classifier of Super-resolution Structural Networks (CLaSSiNet), a conceptually novel computational framework that overcomes these sparsity and heterogeneity constraints. By integrating connectivity, 1D periodicity, and 2D regularity classifiers through newly developed algorithms, CLaSSiNet sensitively captures the organizational signatures of imaged networks to automatically segment and map networks with an unprecedented resolution (∼256 nm, reaching the diffraction limit of light). CLaSSiNet uniquely resolves four distinct organizational states (1D periodic, 2D polygonal, disordered, and non-network), providing a robust platform for analyzing SMLM data sets regardless of labeling chemistry. Using CLaSSiNet, we achieve the first spatially resolved, quantitative mapping of organizational heterogeneity in the actin-spectrin membrane-associated periodic skeleton (MPS), a conserved cytoskeletal network located underneath the plasma membrane of animal cells. This analysis reveals previously unrecognized organizational principles for these MPS networks, with ordered 1D and 2D networks enriched at cell edges and junctions, while non-network states dominate the cell body. Furthermore, we uncover a mechanical coupling principle wherein actin stress fibers bias the symmetry and orientation of nearby spectrin lattices, indicating bidirectional coordination between contractile actin bundles and periodic MPS networks. Comparative analysis across diverse cell types highlights the cell-specific "tuning" of these supramolecular design rules. Broadly, CLaSSiNet establishes a principled computational framework for dissecting the nanoscale design rules of complex molecular networks, offering a robust methodology applicable to the wider study of hierarchical biological and bioinspired architectures.