Hierarchical Spatial Pattern Analysis on Neuronal Neighborhood

We developed a robust method to detect & profile injury-caused alterations to brain tissue at the multi-cellular scale:

Analyzed the interaction of multiple cell types rather than a single cell type
Used a single measurement to measure the local neuron neighborhoods by combining the density and distance of neuron-glial influence; these measurements previously could only be used independently
Represented biological cyto-architecture better than existing methods in terms of the proximity level from neighbors to centers, fitting the uneven distribution of cells, and performing robust to distance parameters

We implemented feature selection on intercellular neighborhood analysis by using Sparse Group LASSO and Tree LASSO algorithms

Fitted pre-defined features with intrinsic hierarchical structures
Enabled to extract significant individual features and feature groups simultaneously
Produced stable and efficient feature selection results

Keywords: LASSO, Sparse Tree, Hierarchical Feature, Feature Selection, Neighborhood Analysis