Thread by @KelvinTuong

Bringing V(D)J analysis to the single-cell space is difficult. We continued to developed dandelion to help with this, finding some interesting new biology along the way in this new pre-print! @Cambridge_Uni @sangerinstitute @ClatworthyLab @teichlab biorxiv.org/content/10.110…

One of things i felt strongly about when writing this was to try and acknowledge and respect the immense amount of work the #AIRR and #singlecell fields has put in to make it what it is today. So I came up with this summary illustration to conceptualise how dandelion can help.

We implemented a number of 'improvements' to a typical workflow, discovering a number of new/novel things along the way (orange is what's new in dandelion).

Firstly, @10xGenomics's V(D)J data contains a lot of non-productive contigs and we think this is because of the enrichment strategy, compared to other methods e.g. multiplex PCR. Will come back to this later...

Secondly, we discovered something we are calling 'mulit-J mappers' - contigs with multiple J genes on the same mRNA, which we propose could be due a natural phenomenon from partial splicing. They tend to happen on non-productive contigs...

Thirdly, we leveraged the single-cell information to come up with a new V(D)J feature space. This is inspired from the pseudobulked neighborhood graph approach by @emmamarydann @MDMorgan_abz @MarioniLab and it let's us do things like compare V(D)J usage patterns across nhoods...

And turns out that we can use this new V(D)J feature space and compute a pseudotime trajectory on it! We went with an absorbing Markov chain approach, implemented in Palantir from @dana_peer's lab, and demonstrate its utility on a developing T cell dataset:

What did we find? Well, this new 'gene expression-anchored' V(D)J pseudotime trajectory showed us that T cell fate commitment started earlier than expected, during the double positive T cell stage! You can see the improved resolution just by eye!

This is not unexpected, considering how CD8 and CD4 T cells would interact with MHCI or MHCII complexes respectively. The new V(D)J trajectory also correlated more with the known biology of 'outward' selection of TRAV and TRAJ genes.

Importantly, this new trajectory aligned better with known fate-commitment transcription factors and also brought out new/novel ones! For example, look at the direction of RUNX3 being incorrectly correlated in the pseudotime constructed using gene expression alone.

Remember I said we found a lot of non-productive contigs earlier? It's not just T cells that express them, but including developing B cells and NK/ILC cells!

We then went back to the V(D)J pseudotime implementation finding that they deviate from the T cells early on, suggesting some ILC/NK probably shared a common developmental trajectory with DN T cells and they probably 'aborted' the pathway due to failed TCR rearrangement

Similar to the T cell trajectory, the new alignment brought out some known patterns of transcription factors for NK/ILC development, but also revealed differences between mice and human.

These results help us formulate a new developmental model with which we are exploring further

A lot of this comes from the amazing co-first authors Chenqu Suo and Krzysztof Polanski (both twitterless, both brilliant 🌟). Also want to acknowledge the amazing environment and help from everyone involved! @ClatworthyLab @teichlab @Muzz_Haniffa @roserventotormo @RLindeboom

What's next? Well we haven't exactly optimised the trajectory analysis for antigen-specific situations and BCRs but that's definitely something we will be looking into! In the mean time, please check out the tool at github.com/zktuong/dandel…

Also I linked this to my mastodon account @[e-mail] so it can be seen there too!

Should also add that all code, most if not all data, and notebooks used are publicly available at github.com/zktuong/dandel…

@Cambridge_Uni @sangerinstitute @ClatworthyLab @teichlab In case people are wondering why the half face don't exist in the pre-print, @emmamarydann thought it was creepy...