Developing new single cell omics methods and bench validating all the hypotheses those techniques give us. Opinions expressed are my own.
Joined June 2018
- Tweets2.6K
- Following681
- Followers1.3K
- Likes5.1K
The metric is very similar to Silhouette from my reading. I think figure 5 says it all though. Most of the integration algorithms are the user deciding how similar the results should be. Everyone just needs to multiplex & daisy-chain tech reps across batches...
1
2
124
We looked at freshly prepared mouse heart, fetal brain, and human PBMCs & found similar classes of *transcripts* differentially abundant
1
1
296
We just stained for the markers in human brain & found no evidence of CD19+ mural cells. A full extra paper in our supplement... This problem of "dataset integration" has now hampered CAR-T research for anti-CD19 cells for B-cell malignancies. Standard integration algs are bad...
1
559
We similarly saw that B-cells got merged with fibroblasts with scVI (tabula sapients paper)... WTF??? These are all publish coordinates and clusters - I didn't touch this data at all, so it's what people are doing in practice that is the problem.
Even if you're integrating the same tissue type all assumed to be in a non-diseased state, you can end up merging populations of B-cells with urothelial cells! No idea how MCCA came to this conclusion, but there you go...
This one paper could have been 5 debate: I've run into the "mission creep" during peer review, when trying to satiate reviewers leading to what could have been 5 separate papers. But that's bad for science as well as the simple dissemination of our work 👇
But that's not all - with our new method sub-clustering is safe to do because you're guarded against false discovery of "novel subpopulations." Subclustering our memory CD4 T-cells, we actually found CD25+/FOXP3+/TOX+, consistent with Tregs. All 100% unsupervised!
1
213
Looking at the (# of cells agnostic) PMD standarized residuals, we found some stand-out overabundant cell states. The most different was a Hif1a high subset of classical monocytes 🤔
1
263
This is the correct response. Almost 2 years addressing reviews, and finished now. Waiting on 1 co-author for the 130 page update
nitter.gq/ScienceScottT/status/1…
Working with multiple scRNAseq batches? Having trouble replicating at the bench? Our recent pre-print may show one reason why: (doi.org/10.1101/2021.11.15.4…) Findings complementary to @lpachter & Tara’s work finding low dim projections can be unreliable representations.
1
2
47