large part of the heatmap of select features is blank

[Jiam1ng]

when I draw heatmap after index cluster, it turned out a big blank, is this normal to see?
here is my code:
proj.sce <- SingleCellExperiment(assays = list(normcounts = proj.mat), colData = proj.cellname)
logcounts(proj.sce) <- log2(normcounts(proj.sce)+1)
rowData(proj.sce)$feature_symbol <- rownames(proj.sce)
proj.sce <- selectFeatures(proj.sce, suppress_plot=F)
proj.sce <- indexCluster(proj.sce)
heatmap(as.matrix(metadata(proj.sce)$scmap_cluster_index))

[Jiam1ng]

and even when I project my dataset to itself, the the similarity is low and labels are all 'unassigned', is there any mistake I have made? thank you

[Jiam1ng]

I set the n_feature very few till there is no blank, but still when I project the dataset to itself, most of the cells are 'unassigned'

[wikiselev]

how does your proj.mat look like? what is the output of proj.mat[1:10,1:10]

[Jiam1ng]

@wikiselev thanks for your reply, here is the output:
f1_GCGTCGGAGTTT f1_GCGCACCATCAA f1_AAGTGGTGTTGT f1_ACTTCTCTTAAN
0610005C13RIK 0 0 0 0
0610006L08RIK 0 0 0 0
0610009B22RIK 0 0 0 0
0610009E02RIK 0 0 1 0
0610009L18RIK 0 0 0 0
0610009O20RIK 0 0 0 2
0610010F05RIK 0 1 2 1
0610012G03RIK 0 0 0 0
0610025J13RIK 0 0 0 0
0610030E20RIK 0 0 0 0
f1_TGGTTCCTTTGC f1_CTCGACGCTCCN f1_CAGGTCCAGCCA f1_ACAGTCTATGGC
0610005C13RIK 0 0 0 0
0610006L08RIK 0 0 0 0
0610009B22RIK 0 0 0 0
0610009E02RIK 0 0 0 0
0610009L18RIK 0 0 0 0
0610009O20RIK 0 0 0 0
0610010F05RIK 4 2 1 1
0610012G03RIK 0 0 0 0
0610025J13RIK 0 0 0 0
0610030E20RIK 1 0 1 0
f1_TTAACCGCAATN f1_CGACAACTTCCC
0610005C13RIK 0 0
0610006L08RIK 0 0
0610009B22RIK 0 0
0610009E02RIK 0 0
0610009L18RIK 0 0
0610009O20RIK 0 0
0610010F05RIK 1 0
0610012G03RIK 0 0
0610025J13RIK 0 0
0610030E20RIK 0 0

[wikiselev]

thanks, looks good to me. Hard to think why this happens without looking at your data, though...

[huiyuz]

I have the same issue. I have no idea what happened. It seems associated with the selected feature genes. If I use the top 500 features, I got a small blank region, but if I use the top 1000 features, I got a huge gap of blank.