Internal KCN: Automatic synthesis of astrocytic trees in silico through generative modeling.

Abstract:

Astrocytes are prominent glial cells in the brain that shape the anatomy and function of our neural circuits. They do so thanks to their complex branched morphology, which infiltrates the neural tissue and wraps around key neuronal elements, allowing modulation of neural activity by various biochemical pathways through the sites of contact. Characterizing astrocytes branched anatomy is key to understanding their functions in brain circuits. However, there is currently no systematic framework for such a characterization. We present a preliminary automated pipeline to systematically classify astrocyte anatomy based on the extraction of macro- and micro-features from experimental 3D astrocyte tracing. Macro-features are scalar measurements that quantify cell anatomy globally. Micro-features are instead vectorized by the multiple branching compartments of the cell. Deploying techniques from statistics, machine learning, and topological analysis we mine for patterns of macro- and micro-features that completely describe the astrocyte branching architecture. In turn, we use a generative modeling approach to simulate those patterns de novo and synthesize realistic astrocyte membrane scaffolds.