mixcr 4.7 vs mixcr 3.0.13

[Miriam-Zu]

I have a dataset of tcr rep seq from blood. Some of the samples were processed with SMARTer Human TCR a/b Profiling Kit (We will call this V1) and others were processed with SMARTer Human TCR a/b Profiling Kit v2 (we will call this V2).
When running mixcr 3.0.13, plotting histograms of the log10-transformed clone frequencies showed a similar distribution (right skewed) for all samples in both V1 and V2.
When running mixcr 4.7 on the same set of samples, plotting histograms of the log10-transformed clone frequencies showed the same distribution for all samples in V1 (right skewed) but a very different distribution for samples in V2 (normal).
Why is this? What was upgraded in mixcr 4.7 which makes such a big difference for samples from V2 but not for samples from V1?
Note: all histograms use the same scale for direct comparison.
In addition, the samples in both groups of mixcr versions are shown in the same order.

[mizraelson]

Version 1 of the kit did not include UMIs, while Version 2 does. MiXCR v3 could not handle UMIs and lacked presets. MiXCR v4 supports the SMARTer Human TCR α/β Profiling Kit v2, taking UMIs into account and applying filters and corrections to reduce noise in the data.

[Miriam-Zu]

Thanks.
When I used MiXCR v3 on the SMARTer v2 data, I was aware that v3 doesn't natively support UMIs. So instead of running MiXCR v3 directly, I used ImmuneProfiler, which first runs MIGEC to collapse UMIs, and then MiXCR v3 on the processed data. So UMIs were accounted for in that pipeline.

However, when I processed the same data with MiXCR v4.7, there is a substantial shift in the distribution of clone frequencies (from right-skewed to more normal in v2 samples).

If UMIs were already collapsed in the MIGEC + MiXCR v3 pipeline, why is there still such a large difference in the output between that and MiXCR v4.7?

[mizraelson]

It's hard to tell the exact reason without deep diving into the data and tracking the path from original reads to clones, but MiXCR has a very different algorithm that handles UMI correction and filtering which is more advanced than MiGEC.