Module 13: Data Science for Connectomics

📊 Feature Engineering

Neurons and synapses contain measurable features such as volume, length, branching, and connectivity. Feature selection affects analysis outcomes.

• Quantitative features from segmentation
• Normalizing and encoding categorical data
• Dealing with missing values

📊 Dimensionality Reduction and Clustering

Use PCA, UMAP, and t-SNE to compress high-dimensional data. Cluster to identify putative neuron classes or structural types.

• PCA and variance explained
• Clustering with K-means and DBSCAN
• Visualizing 2D projections

🌟 COMPASS Integration

• Knowledge: Feature representation, unsupervised learning
• Skills: Analysis, visualization, pattern detection
• Character: Comfort with ambiguity, exploration
• Meta-Learning: Iterative tuning and validation

📚 References & Resources

• Petersen et al., 2021. Cell types in mouse cortex revealed by unsupervised analysis of connectomics. Nature.
• McInnes et al., 2018. UMAP: Uniform Manifold Approximation and Projection. JOSS.
• Colab: "Neuron Embedding and Clustering Pipeline"

✅ Assessment

• Extract a feature matrix from a sample connectome
• Use UMAP or t-SNE to project into 2D space
• Label clusters and infer possible biological types