sketchy SingleR annotation of snRNA-seq data

Question

[myoui3122010]

I used singleR to annotate the data, where the ref was common cells between original dataset and the matrix generated by me using Kallisto-bustools.

meta$annotation_cell_class <- orig_metadata$annotation_cell_class[match(meta$cell_barcode, orig_metadata$cell_barcode)]

But singleR assigned most NA cells to one type of cell. Based on expression pattern majority of Non-neuron express less than 2000 gene, should i filter them out as I don't require them. And should I just annotate them based on clusters from original annotation.

Thanks for your time

Left: SingleR, Right: Copied from Original data