Hi, I'm doing an analysis of the immuno repertoire en T cell from different patients and this in several conditions.
How could I compare two sets of TCR sequences (each set are arround 100 000 sequences of 10 to 15 AA) ? Some sequences may include small variation of one or two AA.
My goal is to highlight sequences that are specific to a given condition. My data are from a NGS analysis and I hope that there is a way with R or any other computer tool to find similiraties inside these sequences. Thank you for your help.
Best regards,
G L
There are a lot of classic, motif discovery/search algorithms(MEME/MAST, MODSIDE, etc.). Here is a somewhat old reference page: https://molbiol-tools.ca/Motifs.htm. These may not be of particular use in this case as you have way too many sequences, but perhaps if you can prefilter based on some criteria you could apply those.
There are also specific TCR analysis tools, but since I have not used those before I can only reference them:
DeepTCR (https://www.biorxiv.org/content/10.1101/464107v1)
Immunomap (http://cancerimmunolres.aacrjournals.org/content/6/2/151)
tcR (R package, https://cran.r-project.org/web/packages/tcR/vignettes/tcrvignette.html).
I think DeepTCR may be what you are looking to do.
Best regards,
Stefan
A Gibbs sampler may be of use here. This has bene used for this purpose before.
Using or comparing 100,000 is neither practical nor logical. Also, for NGS data, you need to cluster the sequences and merge same or similar (within certain threshold) sequence together. There are many software available which can serve this job. In the end you will get very few sequences.
I am not sure that Dr Singh's suggestion would work here. Merging the sequences will only make the problem more difficult.
I have had a similar problem and found that a Gibbs sampler solved the problem. I only had about 1000 sequences rather than 100,000 and the run time was not too bad.
David Mitchell
if you think more about Gibbs sampling for DNA sequence motifs, it computes the local multiple alignments, optimize the likelihood and converge to a global maximum "with in a limit", and that limits corresponds to threshold.
Converging for 100,000 sequences could take good amount of time and resources.
How could you say clustering could cause problem, if the defined threshold is 100 % identity?
I may have misunderstood the idea behind clustering
I was under the possibly mistaken impression that another poster suggested concatenating the sequences to reduce the numbers
There may be no need to do all 10,000 in one go
Once the motif identified you can use this to start a search
The Gibbs is very useful when you do not know what to look for in the first palce
Thank you all for your answers. I will try some of your positions and will be back soon enough.
Best regards
- Badges
- Science method