brainfo / denblab_trial

For $\beta$ test of Dnbelab and compared with 10x V3

An overview:

flowchart LR;
    subgraph preprocess
    cr_raw[(cellranger raw)]-.->eD[\emptyDrops\];
    dnbe_raw[(dnbelab raw)]-.->eD;
    cr_raw-->cb[\cellBender\];
    dnbe_raw-->cb;
    eD-.->scb[\scrublet\];
    cb-->scb;
    scb-->cr_filter[(cellranger filtered)];
    scb-->dnbe_filter[(dnbelab filtered)]
    end
    subgraph integrate
    cr_filter-->scVI[\scVI\]
    dnbe_filter-->scVI
    end
    subgraph downstream
    scVI-->scanpy["`cell type
    features analysis
    variance analysis`"]
    end

Loading

Folder structure

/upstream_stats: reports from cellranger count and dnbelab run
/h5_input: h5 matrices from calling cell with EmptyDrops or Cellbender + cellranger count or dnbelab run
/h5_analysis: general qc before and after scrublet

Temporary summary

with the same parameters for EmptyDrops,

• dnbelab retain double the number of nuclei
• but those more retained nuclei are with less gene numbers

Questions

Are these more retained nuclei are good biological samplings from the tissue?
in terms of cell type identity first
This lead to our next step:

Next step

• clustering and viriance analysis
• cell type markers
• integrate matrices from both platforms with scVI