Two changes, but only the first one was made the deafult at the moment:

1) when computing cluster connections in the 2d layout, we were missing normalization to cluster size. This is now improved. It may influence the optimal T_edge and max_degree, in case you were playing with it, otherwise, it is supposed to improve robustness of the layout.

2) code for identifying orphan cells in a given cluster solution was added. But it is still not used by default or connected to the pipeline. The experimental strategy currently used for identfying orphan cells is based on two metrics - first is the fraciton of balanced k_nn edges that connect the cell to its designated cluster (small number is bad), second is the degree by which the cell is expressing the genes enriched in its desiganted cluster. If both values are low, we believe the cell is not compatible with the cluster model and should be put aside. Experiments with the E9 data show few cells are unlinked with clusters. On the other hand experiment with the T cell dataset in the melanoma project showed many cells (6% or so) are problematic, with strong enrichment for cells with low umi count. WE wil consider adding this operation to the pipeline as an optional feature 

