Clarification on umi_count/duplicate_count/consensus_count

[grst]

In the context of single-cell TCR data, I've always been a bit confused about what to put in which field. I've seen in the latest revision of the AIRR standard, a umi_count field has been added to resolve some of this ambiguity.

Just to be sure I got this right:

• umi_count should contain the deduplicated read count (i.e. the number of unique UMIs)
• duplicate_count now remains empty (what is it actually for?)
• consensus_count should contain the raw read count (before UMI deduplication)

Is this correct?

This came up in scverse/scirpy#478

[bcorrie]

I'm glad you asked that question... 8-) Not...

Other might provide a more precise answer, but my understanding is that duplicate_count doesn't apply to single-cell 10X style experiments - see #543 (comment)

It is for use in non-UMI bulk studies if I understand correctly.

The different fields stem from trying to capture the different types of counts one might have with such varying techniques such as bulk and single-cell.

See the associated very long threads as to how this was arrived at. Perhaps we need to update the docs to better reflect this?

[bcorrie]

I am hoping that @javh @scharch might comment on your use of consensus_count and umi_count, it seems right to me but I am not an expert.

[scharch]

Yes, I think this is correct. See #161 (comment) that duplicate_count is intended for use when there are no UMIs in the experimental protocol.

[schristley]

• duplicate_count now remains empty (what is it actually for?)

Exact sequence duplicates, that is, same length and identical nucleotide sequence. Almost used exclusively by pre-processing tools for bulk AIRR-seq. As those duplicate sequences will have identical annotations from (say) IgBlast, it is used as an optimization to speed up the analysis workflow.

[schristley]

A further point. It is important for downstream analysis tools to be aware and use duplicate_count, especially if they are performing counts or statistics that take the number of sequences into account.

[grst]

Got it, thanks everyone!
Scirpy will use the umi_count field by default from the next release on.