OWClusterAnalysis: Cluster-vs.-cluster instead of one-vs.-all

[mstrazar]

Perhaps applies to differential gene expression in general; If we want to find cluster specific genes, we have to compare the expression in cluster A to its most similar cluster. Otherwise, the rest of the clusters will occlude the results and we will get the genes that do not really differentiate cluster A from all of the other clusters.

[mstrazar]

One possible solution is to compute a per-cluster p-value and use the maximum across clusters.

[anupparikh]

I see two use cases:

• I would like to identify genes that differentiate a cluster from all others (marker genes). As an example, I want to label a cluster as a cell type based on the marker genes.

• I want to know differentially expressed genes between two considers in a given cluster. As an example, I want to know differentially expressed genes in before and after treatment in B-cells.