Dears, I am trying to integrate multiple of scRNA public datasets. I defined the cell types using SCEVAN. I got tumor, normal, filtered and unknown cells. The question now, should I remove the filtered cells before the integration step? because they are filtered due to their low quality. Or should I convert them to unknown and let the scANVI integration predict them ? because they are irrelevant cells that SCEVAN failed to classify them.

Thanks in advance!