My paper on Poisson underdispersion in reported Covid-19 cases & deaths is out in Significance. The claim is that underdispersion is a HUGE RED FLAG and suggests misreporting. Paper: rss.onlinelibrary.wiley.com/doi/10.1111/17… Code: github.com/dkobak/covid-u… Figure below highlights 🇷🇺 and 🇺🇦. /1

I’m happy to share our latest preprint introducing the Single Cell Lung Cancer Atlas (LuCA) integrating >1.2M cells from 223 NSCLC patients + 86 controls from 29 datasets. We used it (among other things) to study tissue-resident neutrophils in NSCLC.🧵⬇️ biorxiv.org/content/10.110…

A phenotypic signature that identifies neoantigen-reactive T cells in fresh human lung cancers dlvr.it/SQ74b3

The analysis of single-cell RNA-seq data begins with "normalizing" counts. In a preprint with sina, ingileif & Angel Galvez Merchan, we examine the assumptions and challenges of normalization, benchmark methods, and motivate solutions: biorxiv.org/content/10.110… 🧵 1/

Our single-nucleus cross-tissue cell atlas is out in Science Magazine as part of Human Cell Atlas: science.org/doi/10.1126/sc… w/ Eugene Drokhlyansky, Shankara Anand, Evgenij Fiskin, Ayshwarya Subramanian, François Aguet, Ayellet Segrè, Orit Rozenblatt-Rosen, Kristin Ardlie, Aviv Regev & many others 1/

A big day for life science The Tabula Sapiens, like a Periodic Table of Human Cells, ~500,000 cells analyzed, 24 tissues / organs Science Magazine "a broadly useful reference to deeply understand and explore human biology at cellular resolution" science.org/doi/10.1126/sc…

A final GI rounds to thank IR legend Dr. Mueller. General surgery at MGH is so thankful for all of his help over the last 4 decades! Mass General Surgery MGHIR

Many theoretical works in ML & high-d stats focus on Gaussian data but why should we care? Real data are definitely not Gaussian, amiright? Well, it might not be such a bad assumption, see plot 👇! How is this possible? Turns out there are universality properties in high-d 1/2

Our new study shows that data availability statements are not very useful; 1670 (93%) authors who indicated that data are available on request either did not respond or declined to share their data with us. Journal of Clinical Epidemiology: doi.org/10.1016/j.jcli…

1/4 New commentary on the big data paradox, i.e., the phenomenon whereby as the number of patients enrolled in a study *increases*, the probability that the confidence intervals from that study will include the truth *decreases* 👉 bit.ly/3xoEvP4

Lucy L. Gao, Josh Popp, Alexis Battle, Daniela Witten and I are excited to share our new preprint!(arxiv.org/abs/2207.00554) We introduce “count splitting”, a flexible framework that allows for valid p-values for differential expression across estimated latent variables. (1/8)

I've been reading Andrei Okounkov's short and accessible expository articles about the work of this years' four Fields medallists and they are wonderful. Highly recommended reading! arxiv.org/abs/2207.03867 arxiv.org/abs/2207.03871 arxiv.org/abs/2207.03874 arxiv.org/abs/2207.03875

Ever wondered what image datasets look like if they could be visualized? We have developed a new algorithm for visualization based on contrastive learning. Joint work with Dmitry Kobak and @CellTypist. The full details are available as a preprint arxiv.org/abs/2210.09879 🧵/16

love making making surgical videos. here's a few of my favorites

A very long overdue thread: happy to share preprint led by Sebastian Damrich from Fred Hamprecht's lab. *From t-SNE to UMAP with contrastive learning* arxiv.org/abs/2206.01816 I think we have finally understood the *real* difference between t-SNE and UMAP. It involves NCE! [1/n]

The best way to make an algorithm outperform t-SNE in a benchmark has always been to compare against scikit-learn's implementation. Glad that it is now being modernized. Great work and congratulations on a sweet bug fix by Dmitry Kobak and Weiyi!

As the tradition goes, here is the list of 22 papers I read, learned from, and wished I had been a co-author (in no particular order):

1. Thrilled and proud to announce our latest preprint: Spectra -Supervised discovery of interpretable gene programs from single-cell data. It is a factorization method that really works and is already in intensive use across projects in my own lab --> doi.org/10.1101/2022.1…

One of most important articles I’ve done… showing the noise in clinic BP measurement is large & makes it impossible to track Rx effects; almost useless in evaluating change from 2 clinic visits. Let me explain… ahajournals.org/doi/abs/10.116… Yale School of Medicine Yale Cardiology Circ: CQO

Incredibly proud to see our benchmark of single-cell preprocessing methods finally published 🥳🥳🎉 We show that despite its theoretical limitations, no other transformation consistently outperforms log(y/s+1). All details at nature.com/articles/s4159… and github.com/const-ae/trans…