using Ibex when dataset was not processed with scRepertoire

[cstrlln]

Hello,
this is a great idea, been looking to try something like this and looked like the T cell people were more advanced.

Have a couple of questions:

• I already have an SCE object with integrated vdj data, I have the aa sequences, V gene identity and of course barcode, and I have subsetted so they are all heavy chain only. Is there a way to use this without starting again with scRepertoire?. Looks from getBCR that I could change the name of my colnames in colData for aa, v genes. to match the ones used by scRepertoire.... Would this work? what else would I need to change or make sure to go directly to runIbex with my SCE.

• which dataset was used for the training? Sorry I might have missed it and not very familiar yet with machine learning.

Finally, a suggestion:
Pulling V genes just with grep from GEX data is not ideal as there are a lot of pseudogenes, I would suggest using chromosomal location or the biotype assigned by cellranger, soomething like:
ig_list <- c("IG_C_gene", "IG_C_pseudogene", "IG_D_gene", "IG_D_pseudogene", "IG_J_gene", "IG_LV_gene", "IG_pseudogene", "IG_V_gene", "IG_V_pseudogene")

And then can query into biomart for genes that have that biotype that are also in your dataset.

Carlos

[ncborcherding]

Hey Carlos,

Thanks for reaching out - you're right the single-cell RNA/BCR space is a little sparse - but you should also check out Benisse too.

In terms of questions:

1. Yes you should be able to modify the meta data by matching the names to the getBCR() internal function. Overall the function relies on two columns: 1) CTaa which is the cdr3 amino acid sequence of the BCR, formatted into "Heavy_Light" and 2) *CTgene which is the gene segments used for both chains, formatted like "HV.HD.HJ.HC_LV.LJ.LC", this should be enough to get the pipeline working.
2. Great questions - this is getting clarified in the resubmission of the paper with a comprehensive list of cohorts. But the models were trained on all public single-cell BCR sequences deposited in the Gene Expression Omnibus that were available before November 2022. I am updating the models with additional sources as well for future versions.

Great suggestion - I will add that to the list of changes to implement for the new release!!!

Hope that answers your questions and please let me know if you have any other questions/suggestions as you start using the package.

Nick

[cstrlln]

Thanks Nick, I'll give it a try getting my data to work with ibex.

Another related question: what is the role of the V genes here, how are they used? Are they just kept for referencing? I gather the calculations are based on the aa properties.

Carlos

[JHYSiu]

Hi. Just trying to do something similar now, I have a single tsv (multiple sampleIDs, multiple donors) where each row is a cellbarcode and each column has all the relevant data (cdr3, v_call etc). Clones are already calculated via Immcantation. I'm currently trying to work out what the best way to feed it into IBEX. Thanks! (an example of what the list of data frames input should look like would be incredibly helpful)

[ncborcherding]

@JHYSiu,

With the new version of Ibex on Bioconductor - circumnavigating the scRepertoire recruitment is still pretty similar - you will need a CTgene and a CTaa column in your meta data that is formatted like scRepertoire - so it will need formatted like: Heavy_Light.

At this time there is not plans to support a wider set of formats because of the sheer variability in inputs that can be seen. Let me know if you have any other questions.

Nick