Experience/advice with changing the barcode_max_ed / barcode_min_ed_diff to increase number of tagged reads

[tpjones15]

Ask away!

Hello,

I was wondering if you have any recommendations for trying to increase the number of tagged reads? We currently go from ~95% of reads being full length to only around 60% being tagged. My understanding is that the expectation for this would be closer to ~80%. Aside from increasing the max editing distance between barcodes and the whitelist, do you have any additional recommendations/things I could manually check to see where these lost reads are going?

I'm going to test increasing barcode_max_ed to 3, whilst keeping the barcode_min_ed_diff at 2, with the hope this doesn't let too much noise through.

Any help on this would be much appreciated.

Best,
Tom

[nrhorner]

Hi @tpjones15

There are a few stages where you could be losing these reads. Not finding the barcode in the whitelist is just one place, which you are trying to address Another is the cell count thresholding. What does your knee plot look like? Which version of the workflow are you using?

Can you post the contents of barcode_assignment_summary.tsv

Thanks,

Neil

[tpjones15]

Hi Neil,

I'm using v2.4.1.

An example knee plot is:

and the output for barcode_assignment_summary.tsv for this case:
bc_corrected 8846222.00
bc_no_shortlist_match 354729.00
bc_shortlist_exact_match 101622958.00
bc_shortlist_multiple_hits 77986581.00
cell_count_read_threshold 2557.00
fraction_of_high_q_reads_in_cells 0.60
high_q_reads_in_cells 94169565.00
high_quality_barcodes 965625.00
high_quality_reads 158076073.00
n_cells_after_filtering 5331.00

The expected number of cells is 5,000 so I don't think being more permissive here would be the way to go.

Thanks,
Tom

[nrhorner]

Hi @tpjones15

Apologies, but could you please run with the latest version (-resume -r 3.0.1) and resend the summary? One of the values in the the bc summary file in v2.4.1 was being miscalculated.

The knee plot may indicate some issues with the data. There does not appear to be a defined "cliff" which we would normally expect to see https://www.10xgenomics.com/support/software/cell-ranger/latest/advanced/cr-barcode-rank-plot.