Lior Pachter Profile picture
Mar 10, 2020 15 tweets 7 min read Read on X
In a new preprint, @sinabooeshaghi et al. present deep SMART-Seq, @10xGenomics and MERFISH #scRNAseq (37,925,526,323 reads, 344,256 cells) from the mouse primary motor cortex, demonstrating the benefits of cross-platform isoform-level analysis. biorxiv.org/content/10.110… 1/15 Image
We produce an isoform atlas and identify isoform markers for classes, subclasses and clusters of cells across all layers of the primary motor cortex. 2/15 ImageImage
Isoform-level results are facilitated by kallisto isoform-level quantification of the SMART-seq data. We show that such EM-based isoform quantification is essential not just for isoform but for gene-level results. #methodsmatter 3/15 Image
Using the 10xv3 data, we also present the first cross-platform validation of SMART-seq isoform quantification (possible by analyzing transcripts with unique sequence near the 3’ end). 4/15 Image
A key result is that many cell types have strong isoform markers that cannot be detected at the gene level. See the example below: 5/15 Image
One exciting application of isoform quantification is spatial extrapolation. We show that in some cases isoform resolution can be achieved spatially even with gene probes. This is a powerful way to leverage SMART-seq for MERFISH and SEQFISH. 6/15 Image
These results argue for a rethinking of current #scRNAseq best practices. We find that SMART-seq complements droplet-based methods and spatial RNA-seq, adding layers of important isoform resolution to cell atlases. 7/15
We recommend droplet-based high-throughput methods for cell type identification, SMART-seq for isoform resolution, and spatial RNA-seq for location information. The whole is great than the sum of the parts. 8/15 Image
The beautiful t-SNE above is a lot cleaner than is usually the case. This is thanks to an idea of @sinabooeshaghi who made it w/ t-SNE of neighborhood component analysis (NCA) dimensionality reduction (@geoffreyhinton, Roweis et al. cs.toronto.edu/~hinton/absps/…) rather than PCA. 9/15
NCA finds a projection that maximizes a stochastic variant of the leave-one-out kNN score given cluster assignments. Intuitively, it projects so as to keep cells from the same cluster near each other, which is exactly what we want. 10/15
You might worry that NCA overfits. It doesn't. We ran a permutation test to confirm that we are seeing real structure in the data. 11/15 Image
As an aside, we found that the t-SNE of the NCA projection preserved global structure better than t-SNE of the PCA projection (e.g., in terms of inhibitory / excitatory neuron classes). It seems that people have been confounding the performance of t-SNE (& UMAP) with PCA. 12/15
The preprint has code associated to every figure, with links directly to Jupyter notebooks hosted on @github (github.com/pachterlab/BYV…). #reproducibility #usability 13/15
All this based on amazing publicly available data from the BICCN, and this is just a preview of the whole mouse brain which is on its way. 14/15 Image
All of this is a result of incredible work by @sinabooeshaghi who did all the analysis single-handedly. Follow him for more interesting #scRNAseq in the near future. 15/15 Image

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Lior Pachter

Lior Pachter Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @lpachter

Feb 21
My blog passed 3 million views today from more than 1.8 million visitors. There have been a total of 119 posts in just over 10 years.
I'm one of those visitors. The blog is an idea repository and I go back sometimes for recall. Some highlights 1/🧵 liorpachter.wordpress.com
Image
Just today I revisited the PCA post to recall some of the properties of the transform. A student, Nick Markarian, taught me the Borel-Kolmogorov paradox today (topic for a future post) and the post was helpful in thinking about some things. 2/ liorpachter.wordpress.com/2014/05/26/wha…
I've been teaching a bit of phylogenetics this year and this post on the Golden-Thompson inequality just came up. 3/liorpachter.wordpress.com/2018/10/05/rat…
Read 24 tweets
Dec 24, 2023
This year I had the privilege of enjoying in-person conferences again, and in April I met @dvir_a & Dan Gorbonos, from whom I learned a bunch of interesting science. Here we are having burgers at Hans im Glück in Bonn.
And now, a 🧵about genocide.. 1/
The topic came up at dinner. History presents a heavy burden for Jews in Bonn.. even 78 years after WWII. The "Hans in luck" restaurant we were dining at is just a few meters from where the local synagogue was burned down during "Kirstallnacht" in 1938. 2/ Image
Although decades have passed since the holocaust, in Bonn the events felt closer in time. We were attending the Bonn Conference on Mathematical Life Sciences, which held a moment of silence for Holocaust Remembrance Day while we were there. 3/
Read 51 tweets
Dec 12, 2023
🌌The virial theorem relates time-averaged kinetic energy of objects to their potential energy.

🧬The Price equation relates change in a trait over time in subpopulations to their fitness.

In we observe that the virial theorem is the Price equation. 1/🧵arxiv.org/abs/2312.06114
The virial theorem is a 150-year old tool in (astro)physics. First described by Rudolf Clausius in 1870 in connection with studies of heat transfer, it gained prominence after it was used by Fred Zwicky in 1933 to posit the existence of dark matter. 2/
The virial theorem is elementary calculus. For objects w/ mass m_1,..,m_n at positions z_1,..,z_n, velocities v_1,..,v_n, & acted on by forces F_1,...F_n, the virial "theorem" is the identity shown below. S = \sum_i p_iz_i (p_i is momentum), U is potential energy; T, kinetic.3/ Image
Read 15 tweets
Dec 11, 2023
This speech by @FareedZakaria is a litany of misinformation. There is much to improve at US universities, but his claims are false and unhelpful.

A rebuttal: 1/🧵
The speech begins with the claim that the public has been losing faith in universities because universities are pushing political agendas instead of catering to excellence. But he provides no evidence of a link.

Hypothesis: the main factor is cost. 2/
foxbusiness.com/economy/colleg…
Image
He says that in 2016 70% headed for college vs 62% now. But provides no evidence that it's because of "American universities abandoning focus on excellence... to pursue... DEI".

America is an outlier among all advanced nations in cost of education. 3/
businessinsider.com/how-much-colle…
Read 25 tweets
Aug 23, 2023
A 🧵 on why Seurat and Scanpy's log fold change calculations are discordant. 1/

(based on the Supplementary Notes from ). biorxiv.org/content/10.110…

Image
I first became aware of the discrepancy in LFC reporting by Seurat and Scanpy from a preprint by @jeffreypullin and @davisjmcc:

The result seemed surprising because it didn't seem like calculating log(x/y) = log(x)-log(y) should be complicated. 2/
So what gives?

We have molecule counts X_{ig} where i ranges over cells & g over genes, and we consider two groups of cells G_1 and G_2 containing n_1 and n_2 cells respectively. Let's start with Seurat which calculates LFC according to the formula below. But what is Y_{ig}? 3/ Image
Read 27 tweets
May 2, 2023
Actually, not transforming the data outperforms log(y/s+1). 1/
The "performance" in this analysis boils down to checking consistency of the kNN graph after transformation. That's certainly a property one can optimize for, but it's by no means the only one. In fact, if it was the only property of interest, one could just not transform. 2/
Of course that is trivial and uninteresting. The purpose of normalization is to remove technical noise and stabilize variance. But then one should check how well that is done. And as it turns out, log(y/s+1) actually removes too much "noise". 3/
Read 6 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(