- Clustering Methods (K-means, PCA, UMAP (?))
- Cell Types
- Morphological features
- Functional Connectivity Rules
- Unsupervised learning
- Dimensionality reduction through PCA
- Overview of clustering methods (scikit-learn)
- K-means clustering (scikit-learn)
- Clustering Hyperparameter tuning (k, initialization, tolerance threshold, etc.)
- Clustering Failure cases
- Cortical cell types
- Data selection/visualization
- Functional features
- Pairwise signal correlations
- Pairwise parameter comparison (i.e. orientation/direction tuning)
How does circuit architecture, i.e. who is connected to who, influence physiological function? Many ways to ask this question.
Like-to-like connectivity. Do cells with similar responses connect more together. Ko et al. Lee et al. etc. (calculate signal correlation, bin pairs of neurons by signal correlation, measure connectivity rates) Add cell types to mix, how does story change?
- Are cells in the same functional clusters more likely to be connected?
- Cluster functional responses, measure connectivity.
- Are cells with similar cell types more functionally similar than cells with different cell types?
- Are cells in the same cell type more likely to be in the same functional cluster?
- Unsupervised clustering based on those pairwise distances
- Ask whether being in the same EM class reduces variance in pairwise functional features
- Ask whether being in the same EM class makes you more/less likely to be in the same functional group.
- Discuss the structural differences between these two ways of asking the question? Why use one versus the other?