Spatial Statistics and Explainable Deep Learning for3D Cell-Protein Interaction Profiling
Spatial Statistics and Explainable Deep Learning for3D Cell-Protein Interaction Profiling
Where did the research take place?
The study site has not been established. Author addresses may differ from where the research occurred.
Publication status: preprint
A plain-language reading has not been prepared for this paper yet.
Original abstract
Abstract Digital histology of human postmortem brain tissue is fundamental to understanding neurodegeneration, yet resolving thespatial interactome of neuroinflammation remains a major challenge due to the limitations of manual scoring and scalargeometric approximations. We present a hybrid computational framework that bridges the gap between mathematical rigorand morphological sensitivity by combining stochastic spatial point processes with explainable 3D deep learning. Usinghigh-resolution spinning-disk confocal immunofluorescence images from a human Parkinson’s disease cohort (Braak Stage3/4), we mapped the spatial relationships between IBA1-positive microglia and phosphorylated α-synuclein (pSyn) aggregatesacross the nigrostriatal pathway. While second-order spatial point process statistics demonstrated that global microglial spatialclustering and macro-distribution remain highly conserved between PD and healthy controls, a voxel-level Overlap Indexrevealed a significant, region-specific elevation of microglial-pSyn co-occupancy exclusively in the substantia nigra (d = 0.74,p = 0.044). To capture the underlying pixel-level transformations, we trained multi-channel 3D ResNet-18 architectures underpatient-contained nested cross-validation, achieving robust classification performance in the substantia nigra (mean AUROC= 0.74, excluding outlier fold) while confirming model integrity through a biological null control in the putamen. Furthermore,human-centred interpretability via 3D Gradient-Weighted Class Activation Mapping and Area of Relevance profiling successfullylocalized the discriminative disease signal to micro-morphological texture gradients along peripheral microglial ramificationsat pathology-rich tissue depths. By mathematically ruling out global density shifts and visually decoding the network’sdecisions, this dual-tiered framework successfully resolves the qualitative morphological transformation of microglia at theneurodegenerative locus, offering a reproducible, explainable paradigm for digital computational neuropathology.