Can we "artificially extend UMIs" and correct for primers mis-binding

[jracle85]

Hello,

I've some TCR-seq data generated through SEQTR protocol (https://www.sciencedirect.com/science/article/pii/S2667237523000784) and I'd like to use MiXCR to analyse this data. I could define some custom preset for it and it works quite well. But there are few subleties related to this protocol and I'd like to know if there are any way to deal with these:

1. Our UMIs are very short (9-mer sequences "H{5}N{4}"), and so the UMI library is not so diverse and we expect the same UMIs to be present in multiple different clones. For the moment, we don't perform the "refineTagsAndSort" step to avoid dropping ~50% reads due to UMI collisions. The UMI sequences are directly preceding V primer sequences (~50 different V primer sequences for TRA), and I would therefore know if we could tell MiXCR to consider for example the real 9-mer UMI sequence + the V primer sequence (or begin of it) as the UMI in its different steps?

   • I know I could just extend the UMI pattern to include these additional nucleotides, but I'd like to keep the V primer sequences in the R1 reads as well for the alignment. So would it be possible to use a part of sequence both in the UMI and R1?
   • Or maybe we could do first a step of alignments and then define the "UMI" as corresponding to the 9-mer sequence + V/J gene names or else?

2. As written above, the protocol has ~50 different V primer sequences for TRA and some of them have only few nt differences. So it happens for example that the primer corresponding to TRAV12-3 gene sequence binds "unspecifically" to TRAV12-2. For this reason, when looking at the aligned reads, we see that some reads have their first ~20 nt corresponding to TRAV12-3 and the subsequent ~30 nt corresponding to TRAV12-2. Is there a way to tell the aligner what are our primer sequences and to tell that it should give more importance in aligning correctly the nt that are not part of the primer sequence?

Thank you very much for your help!

Best regards,

Julien

[mizraelson]

Good thinking! You can do this to get what you need:

^(UMI1:N{9})(R1:(UMI2:N{10})*)\R2:8)

This way, you will have two UMIs, where the first one is excluded from the alignment and the second one remains.

As for the issue with primer misannealing, it’s not really something you can fully prevent. The reality is that some primers anneal to the wrong genes and you actually do have such reads. The clone itself is perfectly valid though. Usually, you would exclude the primer sequence from the alignment, but you mentioned you want to keep them. The question then becomes: what do you want to do with reads where the first 20 nt have 2 mismatches from the primer misannealing, but the rest of the sequence is fine? If you have a sufficiently long region after the primer, the V gene should still be identified correctly. You can also try excluding this segment from the assembling feature (though not by sequence, only by location relative to the anchor points) if you simply don’t want that part of the sequence to be used in assembly and be present in the output.

[jracle85]

Thank you for your answers!

The pattern with R1 containing the UMI2 is what I was looking for, that's great. I just don't understand what is the \R2:8) part? It doesn't work, but it works well when only using the pattern: ^(UMI1:N{9})(R1:(UMI2:N{10})*) (I'm using single-end reads).

And concerning the primer misannealing, it'll then certainly be better to remove these primer sequences from the alignment, thank you.

All the best,

Julien

[mizraelson]

Sorry, that was a typo on my end. I meant R2:*

[jracle85]

Perfect. Thank you.