Discrepancy Between MiXCR CDR3 Sequences and scRNA-seq Data

[RomiGoldner]

I'm encountering an issue while correlating MiXCR and 10x Genomics scRNA-seq data. After running mixcr analyze 10x-sc-5gex, I extracted CDR3 sequences from the nSeqCDR3 column. However, I can't find these sequences in the corresponding RNA-seq data from the same cell barcodes.

Could MiXCR's processing lead to CDR3 sequences that don't match raw RNA-seq reads?
Any advice on how to approach this discrepancy?

Appreciate any insights or suggestions!

[mizraelson]

Hi,
Based on your message, I understand you performed only 5' gene expression analysis and not a V(D)J-enriched library, correct? Where do you look for the CDR3 sequences, is it the raw FASTQ files? How do you identify the CELL barcode? Since MiXCR corrects erroneous barcodes, there's a possibility that a read containing the CDR3 sequence is tagged with a barcode differing from the raw data.

MiXCR processing can't create a CDR3 sequence that was not present in the reads. you can trace the clones back to the original reads if necessary. To do this, follow the instructions below:

First, run the analysis with these additional parameters:

mixcr analyze 10x-sc-5gex \
-Malign.parameters.saveOriginalReads=true \
--prepend-export-alignments-field -descrsR1
--prepend-export-alignments-field -descrsR2
input_R1.fastq.gz \
input_R2.fastq.gz \
sampleName

Then, export alignments from the .vdjca file generated by the above command:

mixcr exportAlignments \
sampleName.vdjca \
sampleName.alignments.tsv

In the sampleName.alignments.tsv file, you'll find alignments for each successfully aligned read, including headers for these reads. Note that due to fragmentation, not all reads will encompass the CDR3 region. These reads can then be located in the raw FASTQ files by their headers. Not all of these alignments will end up being assembled to clones, but you can manually check that each CDR3 sequence comes from a certain read.

Let me know if it helps.

[RomiGoldner]

MiXCR v4.6.0

[mizraelson]

So I suppose have you already run:

mixcr analyze 10x-sc-5gex \
-Malign.parameters.saveOriginalReads=true \
input_R1.fastq.gz \
input_R2.fastq.gz \
sampleName

Can you try running after:

mixcr exportAlignments \
    -descrsR2 \
    sampleName.vdjca \
    sampleName.alignments.tsv

[RomiGoldner]

Thank you, that worked.

targetSequences: AAACAGAGTTTGTTTCATATCACAGCCCCCCAGCCTGCAGACTCAGCTACCTACCTCTGTGCACGGAACACCGGTAACCAGTTCTATTTT

nSeqCDR3:
TGTGCACGGAACACCGGTAACCAGTTCTATTTT

descrsR2:
A00519:644:HCNM3DSXY:1:1576:30942:23265 2:N:0:CACCGCACCA+GACTGTCAAT

FASTQ (R2):
@A00519:644:HCNM3DSXY:1:1576:30942:23265 2:N:0:CACCGCACCA+GACTGTCAAT
AAAATAGAACTGGTTACCGGTGTTCCGTGCACAGAGGTAGGTAGCTGAGTCTGCAGGCTGGGGGGCTGTGATATGAAACAAACTCTGTTT

[mizraelson]

So, the R2 reads for 10x are in reverse complement orientation, as shown in the image.

However, the sequences of CDR3 and other regions in MiXCR output are presented in the direct orientation of the coding sequence.

By reverse complementing R2 from your file, we can observe the sequences as follows:

reverseR2: AAACAGAGTTTGTTTCATATCACAGCCCCCCAGCCTGCAGACTCAGCTACCTACCTCTGTGCACGGAACACCGGTAACCAGTTCTATTTT
                                                                    |||||||||||||||||||||||||||||||||
nSeqCDR3:                                                           TGTGCACGGAACACCGGTAACCAGTTCTATTTT

[RomiGoldner]

Thank you very very much for your help.
This helped a lot.