Fewer barcode assignments when specifying linker sequence instead of linker length

[graceoualline]

Hello,

I noticed that specifying the linker sequence, instead of the linker_length in the config results in fewer barcode assignments in the output.

After some testing, I think this happens because when the linker sequence is provided, the pipeline uses Cutadapt to extract barcodes rather than trimming by fixed length. This produces a barcode file where barcodes can vary in size, and any barcode that doesn’t match the bc_length specified in the config file gets filtered out later in the pipeline.

If the barcode is always the first "N" bases, it seems better to specify only the linker_length instead of the linker sequence in the assignment config. That way, the pipeline extracts barcodes purely by length, avoiding Cutadapt’s variable trimming behavior.

Example:
On a small test set (40,000 reads):

• When specifying the linker sequence, I got 33,994 assignments.
• When specifying only the linker length, I got 34,564 assignments.

Suggestion:
I wonder if it might be helpful to clarify in the documentation that providing a linker sequence triggers Cutadapt-based barcode extraction, which can sometimes result in fewer assignments. Specifying only the linker length uses strict position-based extraction and may be more appropriate when the barcode is at a fixed position.

Thanks so much for all the work on this pipeline!
Grace

[visze]

Hi Grace,

Thank you for your comment and suggestion. I will extend the documentation to clarify this point.

I agree that I see fewer BCs when using linker instead of linker_length, for the reason you described. However, in the datasets at hand, this choice influenced the replicate correlation. Using cutadapt resulted in higher replicate correlation. My explanation is that this approach "filters out" reads—and therefore barcodes—that are erroneous, leading to a lower false assignment rate of BCs to oligos.

However, I expect this can vary across different experiments. From the data I have analyzed, I would recommend using linker. But these are just a few, and our data usually has a separate index read for the BC.

Best,
Max