Issues about similarities in reference cells

[liuhao-stark]

Hi there,
Thank your for developed such an useful tool to make cellular annotation more faster and easier. @mhemberg
Recently, I'm trying to utilize scmap to calculate the similarities between different timepoints of my scRNA data. For instance, I want to compare the transcriptomic similarites of my 3 month Myeloid cells with the same cell type at baseline. Because I used 4000 features as variable features in Seurat process. Therefore, I used 4000 again in setFeatures() during scmap analysis. I picked baseline cells as 'reference' and cells from samples at 3 month as 'query' , something confusing emerged:
I want use the scmap similarites between reference cells and query cells to quantify the difference between two timepoints, so after I splited "Culture_Map$sce_All$similarities " by the prefix of cells like 'query' or 'reference' , I got two matries. one is about similarites in reference cells it self, another is about similarites between reference and query cells, I could found some similarites in the first matrix is lower than the second one like this:

mean(p3_query_matrix)
[1] 0.6334649
mean(p3_reference_matrix)
[1] 0.6198853
mean(p4_query_matrix)
[1] 0.6703887
mean(p4_reference_matrix)
[1] 0.6323424
I could not understand why this happened, could you give me some advises? Any advises is welcome.
Thank you!