SyConn

SyConn infers synaptic connectivity matrices and classifies neuronal and subcellular structures from teravoxel-scale volume electron microscopy (EM) datasets to enable large-scale neural circuit analysis.

Key Features:

• Machine learning methods: Uses deep convolutional neural networks and random forest classifiers to detect and classify synapses and subcellular structures.
• Connectivity inference: Automates generation of synaptic connectivity matrices from volume EM data.
• Skeleton integration: Processes manual neurite skeleton reconstructions to assign synaptic contacts to neurites.
• Structural classification: Identifies and classifies mitochondria, synapses (including synapse types), axons, dendrites, spines, myelin, somata, and cell types.
• Scalability: Operates on teravoxel-scale volume electron microscopy datasets, including serial block-face EM.

Scientific Applications:

• Cross-species EM connectomics: Applied to serial block-face electron microscopy datasets from zebrafish, mice, and zebra finches for large-scale wiring analysis.
• Songbird basal ganglia wiring: Computed synaptic wiring of songbird basal ganglia to relate connectivity patterns with physiological data.
• Physiology–ultrastructure correlation: Enabled discovery of correlations between high in vivo firing rates of basal-ganglia cell types and increased densities of mitochondria and vesicles.
• Synapse scaling analysis: Revealed systematic scaling of synapse sizes and numbers depending on the postsynaptic cell types they innervate.

Methodology:

Applies deep convolutional neural networks and random forest classifiers to volume EM data and manual neurite skeleton reconstructions to infer synaptic connectivity and classify organelles, synapse types, and cell types.