Custom clustering unexpected behavior

[DenisSch]

By Zach Avigan:

[...] I am running into difficulty trying what you mentioned with custom clustering. I tried merging multiple phenograph clusters together (merging different clusters that actually made up the same cell type). However, when I perform custom clustering on this, the results are somewhat erratic.

For example, say I merged cluster 1-3 out of 22 into a manual cluster and then run custom clustering on this. Sometimes, it simply returns all 22 phenograph clusters in reverse order (the numbers associated with each one match the amount of cells in each original phenograph cluster, in reverse order). Other times, it only returns one custom cluster with the same number of cells as the sum of clusters 1-3. Additionally, when experimenting with different manual clusters, I sometimes found that it would output the "correct" output from the custom clustering I had tried 5-6 attempts ago. I had the same issue when trying with manually gated cells on a tSNE plot.

A few times, it has worked as expected (i.e. returning one custom cluster numbered "100" and then the remaining pheno clusters 4-22). However, when I run a neighborhood analysis on this clustering, it simply gives the pop-up window that there is no clustergram with only one image selected, but it does not output a heatmap graph of the interactions.