A cell atlas is a collection of annotated cell types, clustered by similar gene expression. The proximity of the cell types as shown by the projection infer lineage, meaning one cell type begat another. By taking samples at various time points, you can unearth more and more (rare) cell types transitioning from cell type to another, giving you lineage specification from stem cells all the way down to the mature cell types of your tissue of choice. Here they went out on a limb and went with limb.
Loads: blood, brain, (mouse) gastruloids, and many others.
It's not as exciting as a breakthrough as it sounds though. One lab was rich enough to buy the existing tech to sequence it all, and the analytical pipelines are mature enough to do the clustering and lineage tracing for you. Overcoming the batch effects and integrating the different data into each other is still a legitimate struggle, so props to them on that.
Essentially, this a "FIRST!" kind of post, without much science behind it, but it does allow others to tease out new development pathways from the same data, and it acts as a control dataset for others who study mutants which usually have a gene or two knocked out.
The knockouts are where the real science is, but I'm possibly gatekeeping in those last two paragraphs there.
I'd say you're pushing from flippant to disrespectful on the who and how of this being done. Starting with you've got Addenbrooks and Guangzhou hospitals collecting the foetal material that will have been lovingly donated from some of the most distressed members of the British and Chinese population. The pipeline although commodity came from a new startup Enhanc3D Genomics Ltd (inc. 2020). Over 15 labs worked collaboratively to build the atlas of results, because the expense in human time alone far exceeds what could be done in a single lab.
I am being flippant I suppose, but it is also so incredibly easy to publish such works when you have the money of an entire consortium behind you.
Also, how are these datasets to be shared? In some of the sets I've worked with, I've had to reintegrate subsets myself because though the authors published the embeddings of their data (e.g. cell1 is at position X and Y on PC1 and PC2), they flat out refused to publish the loadings (e.g. PC1 is made up of GeneX, GeneY, ...) making it obscenely hard to project your own data on to these reference Atlas's.
Yes, I'm disillusioned in general when I see these papers.
