How to normalize TCR-seq data from different platforms and generated by different methods?

[linqy-immune]

Hi,
I'm here again.😊 I am still a big fan of mixcr and I encounter such a problem mentioned in https://www.biostars.org/p/365321/ which claiming that TCR-seq data could be normalized by mixcr-downsample.
There are several datesets originated from different technologies and platforms like Adaptive Biotechnologies、Illumina、Ion Torrent PGM. The sequence depth is different varying from each other so the abundace analysis seems very comparable. But I am afraid that such difference is caused by differnt sequence methods and platforms. What can I do to intergrate these datesets or normalize these datesets? Could you please give me some suggestions and instructions on how to normalize TCR-seq data? I really need your help!!!
Looking forward to your reply.
Thanks and best regards,
Linqy

[mizraelson]

It all depends on the structure of the library you have. If all the data includes UMIs, it’s better to normalize by an equal number of UMIs. If that’s not available, normalizing by an equal number of reads should work. For some purposes, using the top number of clonotypes might be more relevant.

[linqy-immune]

Hi @mizraelson
Thank you so much for your advision. The datesets are generated by multiplex PCR and 5'UMI-tagged 5′-RACE respectively. I wonder whether it is feasible to integrate and how I can normalize them? I am sorry I didn't get what you mean on how to normalize them above😭😭😭
There is another question not relevant to this topic, what preset should I use when analyzing a dataset whose study design was "Design: According to the manufactures introduction, Single-cell RNA-seq libraries were constructed using Single Cell 5 Library and Gel Bead Kit, Single Cell V(D)J Enrichment Kit,Human T Cell(1000005) and Single Cell V(D)J Enrichment Kit, Human B Cell (1000047)." I tried the 10x-sc-xcr-vdj preset, but the processed result seems not qualified cause there are warnings: Successfully aligned reads: 75.09% [WARN] Unassigned alignments in clonotype assembly: 90.87% [ALERT] Alignments dropped due to low sequence quality: 0.0% [OK] Reads dropped in tags error correction and filtering: 5.62% [OK] CELLs artificial diversity eliminated: 0.0% [OK] Reads dropped in tags filtering: 5.62% [OK] Cell barcodes used in result groups: 76.36% [ALERT] . Can I process these data by mixcr? Could you please give me some tips?
Thans and best regards,
Linqy

[mizraelson]

1. If only part of your data has UMIs but you want to compare all samples, you should normalize by the number of reads.
2. Was this data obtained using a 10x Genomics kit? If not, you would need to know the library structure (e.g., the location of cell and UMI barcodes, etc.).

[linqy-immune]

Hello!
Thanks for you patience. It is very kind of you!

1. You mention that I need to normalize the data by the number of reads, do I need to interagate all the data together into a file?
2. The information about this dataset library is attached here, could you please help me have a look? Which preset should I use?
   
   Thanks and best regards again,
   Linqy

[mizraelson]

1. You can pass multiple clns files to mixcr downsample.
2. This looks like a 10x protocol, so the preset is 10x-sc-xcr-vdj

[linqy-immune]

Oops, thanks so much for your suggestions. They are very helpful. I would give them a try.
As for the second question, I have already tried the 10x-sc-xcr-vdj preset, but the processed result seems not very good cause there are so many warnings: Successfully aligned reads: 75.09% [WARN] Unassigned alignments in clonotype assembly: 90.87% [ALERT] Alignments dropped due to low sequence quality: 0.0% [OK] Reads dropped in tags error correction and filtering: 5.62% [OK] CELLs artificial diversity eliminated:0.0% [OK] Reads dropped in tags filtering: 5.62% [OK] Cell barcodes used in result groups: 76.36% [ALERT]
Does it make sense for me to do some down-sample analysis by these results?
Thank you again for your professional answer and patience.

[mizraelson]

Most likely, the warnings are related to the quality of the data. It all depends on what you are interested in, but usually, single-cell and bulk data are analyzed separately, unless you want to do a specific analysis like building SHM trees from a combined bulk and single-cell dataset.

[linqy-immune]

Thanks for your quick reply, this really helps me a lot and inspires me to proceed with my bioinformatics analysis.
As a matter of fact, I don't plan to integrate bulk and single-cell datasets together. I am just kind of worried that the low quality data may lead to discredible results. Are there any methods to improve the data quality?
It is so kind of you and I trully appreciate your help!

[mizraelson]

Well the data quality relies completely on the wet lab part of the experiment. GIGO principle applies here as usual! :)

[linqy-immune]

lol, that's it. Maybe I need to consider whether recruiting them into my analyzing dateset.
Anyway, thanks a lot for your kind help.😊