How to interpret the Trex result?

[Lucyyang1991]

Thanks for the effort to develop this package! After runing Trex with TCR alpha and beta chain, I integrate them with RNA result. Finally I got a wnn UMAP, I'm just wondering can I assume that the TCR sequence of the same wnn subgroup can recognize the same epitope? How can I prove that? Or do you have any more offical interpretation of the result?

sc <- runTrex(
  sc,
  chains = "TRA",
  encoder.model = "VAE",
  encoder.input = "both", #error
  reduction.name = "TRA.both"
) # take long time

sc <- runTrex(
  sc,
  chains = "TRB",
  encoder.model = "VAE",
  encoder.input = "both",
  reduction.name = "TRB.both"

sc <- FindMultiModalNeighbors(
  sc,
  reduction.list = list("pca", "TRA.both", "TRB.both"),
  dims.list = list(1:30, 1:10, 1:10),
  modality.weight.name = "RNA.weight"
)

sc <- RunUMAP(
  sc,
  nn.name = "weighted.nn",
  reduction.name = "wnn.umap",
  reduction.key = "wnnUMAP_"
)

sc <- FindClusters(
  sc,
  graph.name = "wsnn",
  resolution = seq(.1, 1, .1),
  algorithm = 3,
  verbose = FALSE
)

[ncborcherding]

Hey @Lucyyang1991,

Good question - I am not sure you can say that the closely related cells from the trimodal embedding are truly "epitope-specific" without showing possibly previously established epitope mapping (like using VDJdb or annotateDB()) or actual in vitro work.

In our own work, we found that although closely related in terms of position and clustering, the clones we infered spike antigen specificity mapped to different epitopes along the spike protein. I think this makes sense as we are including RNA into the reduction, so you are getting a "reactive" expression profile.

We are working on some newer models for true "epitope" specificity - so watch this space, but for now, I think "antigen" or "reactive" might be better terms.

Nick