I have 12 rat samples that I'm processing in Seurat. The problem is they look really different from each other in QC, the sequencing depth, cell counts, and the nFeature/nCount distributions vary a lot between samples (violin plot attached). Some samples sit high (median ~3000 features), some are clearly lower quality (median closer to 1000), and a couple are very low.
They were also sequenced separately at different times. There are two timepoints spread across 4 batches, and to complicate things one of those batches has both timepoints sequenced together.
My main question: is it OK to set QC cutoffs (upper and lower limits for nFeature_RNA and nCount_RNA) separately for each sample instead of using one cutoff across everything? Or is that considered bad practice / does it introduce bias somewhere?
I would also integrate them all together after QC to proceed with my analysis, would it be considered okay to run integration after samples are QCed seperately?
1 answer
I even do per sample and celltype. Can matter a lot, for example a macrophage or stromal cell can well have like 5000-7000 detected genes while a neutrophil (since transcriptionally comparably low) only has 1000-1500. Do as granular as you can in terms of biology if this applies. Metrics can change between samples depending on quality.
Log in to answer this question.
