Inconsistent FinalClonotypes between two sequencing platforms on the same sample

[xin-bang]

Hi MiXCR developers,

I noticed that when processing the same biological sample sequenced on two different platforms, the FinalClonotypes count (number of distinct clonotypes) differs significantly between the two results, even though the processing pipeline and parameters are the same.
This leads to concerns about how platform-specific differences affect MiXCR results, especially regarding UMI handling and clone assembly.

Questions / Clarifications Needed:

1. Could MiXCR internals explain this difference?
2. Since I’m using the same tag pattern, library, and features, is there any MiXCR parameter recommended for cross-platform consistency?
3. How does MiXCR handle low-frequency UMIs or borderline alignments during clonotype assembly?
4. Are there any best practices for assessing whether the difference is biological or technical (e.g., comparing CDR3 spectrum)?

Thanks in advance for your help!

Here is more detailed information

1: Despite using identical pipelines and parameters, the FinalClonotypes count differs significantly between platform 1 and platform

<style> </style>

Batch	Sample	TotalReads	AlignedReads	Overlapped	FinalClonotypes	TRA	TRB	TRG	IGH	IGK	IGL
Platform1	JXSCD16_TCR_251228_subseq	12871959	6628998(51.5%)	3153316(24.5%)	196902	80664	107357	6182
Platform1	JXSCD17_TCR_251228_subseq	12871959	6968575(54.14%)	3129433(24.31%)	103627	45942	54515	2638
Platform1	JXSCD19_TCR_251228_subseq	12871959	6303977(48.97%)	3964185(30.8%)	229681	97328	125793	4939	1
Platform2	JXCSD16_TCR_251228_subseq	12871959	6098187(47.38%)	1721450(13.37%)	106918	51833	50417	2646
Platform2	JXCSD17_TCR_251228_subseq	12871959	4995587(38.81%)	2617537(20.34%)	57192	30383	25365	1077
Platform2	JXCSD19_TCR_251228_subseq	12871959	5316726(41.3%)	2563480(19.92%)	103200	54750	45117	2193			D

2: I used the following MiXCR pipeline for both datasets:

Align with UMI tag pattern
mixcr align -t 8 --species hsa --assemble-clonotypes-by CDR3 --library imgt.202312-3.sv8 --rna -f -p generic-amplicon-with-umi --tag-pattern "^N{0:2}aagcagtggtatcaacgcagagt(UMI:N{14})tcttgggg(R1:*) \ ^(R2:*) || ^(R1:*) ^N{0:2}aagcagtggtatcaacgcagagt(UMI:N{14})tcttgggg(R2:*)" --floating-left-alignment-boundary --floating-right-alignment-boundary c --report ${sample}.align.report.txt --json-report ${sample}.align.report.json ${fq1} ${fq2} ${sample}.vdjca

RefineTagsAndSort
mixcr -Xmx40g refineTagsAndSort --report ${sample}.refineTags.report.txt ${sample}.vdjca ${sample}.refined.vdjca

Assemble clonotypes
mixcr assemble --assemble-clonotypes-by CDR3 -OassemblingFeatures="CDR3" -OseparateByV=true -OseparateByJ=true -OseparateByC=true --report ${sample}.assemble.report.txt ${sample}.refined.vdjca ${sample}.clns

Export clones
mixcr exportClones ${sample}.clns ${sample}.clones.tsv

[mizraelson]

Hi, there is nothing in MiXCR that will introduce bias between the sequencing platforms. It is possible however if the sequencing quality was different you got different yield.

[xin-bang]

Hi, thank you for the reply!

I understand the differences are unlikely due to MiXCR itself, but I would like to better understand the platform-specific discrepancies in results and clarify some questions regarding the software metrics and reports.

To help clarify the technical root cause, could you please help answer a few more specific questions:

1. UMI Extraction / Collapse Logic

• How does MiXCR handle UMIs with low read support (e.g., singletons)?
• Could differences in UMI complexity / sequencing error profiles lead to different effective clonotype counts? I'm not sure if this is the reason

2. Clonotype Assembly Logic
When I use:

  mixcr exportClones \
    -aaFeatureImputed VDJRegion \
    -nFeatureImputed VDJRegion \
    my_clones.clns \
    clone_sequences.tsv

My understanding is that nSeqImputedVDJRegion represents the nucleotide sequence of the full V(D)J region assembled from the reads/UMIs. However, when I map my reads back to this imputed VDJ region, I often see that only part of this imputed VDJ region is supported by actual read coverage. This is a bit strange

Additionally, when using mixcr clonesDiff --use-v --use-j --use-c -f --print-clones to compare the two platforms, many clonotypes reported as UniqueIn1 (platform1) can still be found in platform2 . However, when comparing the full nSeqImputedVDJRegion, nucleotide differences are observed in the regions flanking CDR3 (upstream/downstream), which likely causes these clones to be treated as distinct clonotypes. Is my understanding correct?

3. Reports for Debugging
I want to identify diagnostic metrics in the MiXCR reports that might help pinpoint why two sequencing platforms produce different FinalClonotypes counts for the same sample.

If it helps, I included the reports here:

• Two align.report.json (one per platform)
• Two assemble.report.txt

Platform1_JXSCD19_TCR_251228_subseq.align.report.json
Platform1_JXSCD19_TCR_251228_subseq.align.report.txt
Platform1_JXSCD19_TCR_251228_subseq.assemble.report.txt
Platform1_JXSCD19_TCR_251228_subseq.refineTags.report.txt
Platform2_JXCSD19_TCR_251228_L03.align.report.json
Platform2_JXCSD19_TCR_251228_L03.align.report.txt
Platform2_JXCSD19_TCR_251228_L03.assemble.report.txt
Platform2_JXCSD19_TCR_251228_L03.refineTags.report.txt

Thank you very much for your help—your guidance will help ensure these differences are biological rather than technical.