Kristina15
Putative synapse locations were obtained by Anish Simhal.
Around each synapse across each channel the sum of pixel intensities within an 11x11x11 cube are calculated to produce a feature vector. These feature vectors are combined into a data matrix.
The feature columns of the data matrix are then z-scored before running the matrix through meda.
We did this for samples of size 1,000 and 10,000.
The locations used were from the top half of the entire dataset as the rest of the locations were still being downloaded.
The sampled locations are avialable to view in the whole volume can be found here (they are large, > 800 MB): sample 1e3, sample 1e4
Sample size 1,000 (Left) and 10,000 (Right)
Outliers as given by randomForest
Cumulative Variance with Elbows
Hierarchical GMM Classifications
Hierarchical GMM Dendrogram
Stacked Cluster Means plot
Cluster Means