the prediction.score.max was low after joint scRNAseq and scATACseq

[Shuresearcher]

I have two bone marrow LSK samples (treatment and control), merged followed by quality control, normalization, and gene activity matrix computation according to the pipeline in signac website. Than I did cell type annotation using my scRNAseq data according to “joint RNA and ATAC analysis methods (based on 10x multiomics guidelines)”. My scATACseq and scRNAseq are from the same cell source but were obtained from two independent experiments. But the prediction.score.max was lower than 0.5 in the most of my cell type. The lower prediction scores(<0.5) indicated a low confidence in the assigned cell identity. I tried to adjust the parameters for dimensionality reducation , normalization and scaling, but the prediction.score.max was not change. My samples are hematopoietic stem cells and progenitor cells. I think maybe because they are closely related. But I am not sure. Can I believe this cell type annotation for scATACseq? Can I improve prediction scores? If there is another method I can try to make cell type annotation better? Any suggestions will be very helpful for me. Thank you very very much!