1 Synaptome Statistics: Notes

Putative synapse locations have been detected in K15 with Forrest’s synapse detection algorithm and in W with Anish’s synapse detection algorithm. For each feature channel (Synapsin, VGlut, psd95, etc.) an 11x11x11 cube is extracted around each each putative synapse location and the voxel values are summed, creating a feature vector of length (number of channels). This gives us an \(n \times d\) matrix, where the \(n\) rows correspond to putative synapses and the \(d\) columns correspond to the summed immunoflorescence in each channel.

1.1 Clustering

We have implemented our own Hierarchical Mclust function by augmenting Mclust. In the course of exploring we used the full suite of models available in mclustModelNames p. 88

After looking through the BIC plots of each of the 11 models for each node of the tree it seemed best to use the unconstrained model “VVV” = ellipsoidal, varying volume, shape, and orientation.

1.2 MEDA Run on Z-scored data

1.2.1 Cumulative Variance with Elbows

1.2.2 1-d Heatmap

1.2.3 Correlation Matrix

1.2.4 Location plots

1.2.5 Outliers as given by randomForest

1.2.6 Paired Hex-binned plot

1.2.7 Hierarchical GMM Classifications

1.2.8 Hierarchical GMM Dendrogram

1.2.9 Stacked Cluster Means plot

1.2.10 Cluster Means

2 Cluster Synaptograms

The below table contains links to synaptograms from each cluster as given above. The 5 synapses with feature vectors closest to their cluster mean were chosen as representatives.

C1111 C1112 C1121 C1122 C1211 C1212 C122 C21111 C21112 C2112 C212 C2211 C2212 C222
1 1 1 1 1 1 1 1 1 1 1 1 1 1
2 2 2 2 2 2 2 2 2 2 2 2 2 2
3 3 3 3 3 3 3 3 3 3 3 3 3 3
4 4 4 4 4 4 4 4 4 4 4 4 4 4
5 5 5 5 5 5 5 5 5 5 5 5 5 5

2.0.1 The following is a synaptogram from cluster C1121 @ x=770, y=6473, z=17

And as you can see, both Synapsin channels and both VGlut1 channels are “hot” which match the cluster mean as given above.