Posts on X

Mars got a little bit closer today - congrats @SpaceX for landing Starship after 10k flight and belly flop maneuver! 🚀😎

Interesting argument from @MohieddinJafari et al on replacing the term "experimental validation" in computational papers with "experimental corroboration" to avoid connotation of legitimization. Fun read: genomebiology.biomedcentral.com/articles/10.11…

Happy that our single-cell meta-analysis of SARS-CoV-2 entry genes is out @NatureMedicine. Headed by Chris Muus & @MDLuecken in a big effort of the whole @humancellatlas Lung Network, we find age&sex-dependent cell-type specific effects across 31 studies. nature.com/articles/s4159…

Cool: #AI has conquered 'Montezuma's Revenge', one of the harder game problems for deep reinforcement learning so far, out @nature today. Simple solution, remember promising places you've been before. Brings back memories, played this game as 10-year old. nature.com/articles/s4158…

Germany and digitalization - still some todos I guess: For example our vaccination registration system requiring separate emails for everyone raises problems - so I was wondering about pros and cons of the government simply issuing an email account for every citizen?

Excited to share 'Squidpy', our toolbox for spatial single cell analysis. Led by @g_palla1 & @HannahSpitzer1, we extend #Scanpy's infrastructure and analyses to store, manipulate & interactively visualize spatial omics data. Read at biorxiv.org/content/10.110… squidpy.readthedocs.io

Deep learning of the peptide universe from 1M experimental values, out in @NatureComms today! In a great collab with @labs_mann, we train an interpretable recurrent neural network to predict a peptide's collisional cross section&predict it for unseen ones. nature.com/articles/s4146…

I'll be going on partial parental leave next Monday, so please excuse my upcoming partial inactivity - or my focus on all things around babies, prams and baby sport or so. Be back in full in September - take care 😊

Honored to be part of the #ELLIS network: we're currently going through PhD recruiting, and I find it so fun that some of us have actually come up with a novel student-PI matching approach: arxiv.org/abs/2101.12080 More at ellis.eu and with our @ELLISmunichunit

Made my day: In a chat about cycling and why riders only achieve top performance a bit later in life even though strength may decrease, John @MarioniLab made the thoughtful comparison to a scientist's life. It's all about the pain threshold 🧐

Great talk by Barbara @TreutleinLab at our biweekly @singlecellomics seminar on human organoid development, with a very impressive vision towards a Human Organoid Atlas within @humancellatlas activities. Big experimental but also fun computational challenges!

Am I doing too many VCs when my AirPod batteries keep running out before I get up and manage to recharge? 🧐

Congratulations #USA for this new administration - great hopes & expectations all around the world!

Interesting but somewhat scary Sunday read from @wired about how #AI (GPT2) was used to submit auto-generated comments to a US government website - this shows how hard it will be in a largely digital democracy to judge public opinion. wired.com/story/ai-power…

Happy that transfer VAE, a generative model to predict out-of-distribution single-cell perturbations such as drugs or disease is out in @OUPBioinfo. Led by @MohammadLotfol1 and @falexwolf, we formulate response prediction as a style transfer problem.
academic.oup.com/bioinformatics…

Single cell proteomics finally becoming a reality, with awesome miniaturization effort by @labs_mann! Robust plate-based readout (>1200 proteins), and strong differences to RNA; we (@_sabrinarichter @davidsebfischer ) contributed adapted #scanpy analyses. biorxiv.org/content/10.110…

Excited to share our data&model zoo sfaira: We provide 233 data sets, 55 organs, 3.1M annotated cells together with 8 pre-trained model classes, to replace manual workflows, facilitate method benchmarking & extension w private data. biorxiv.org/content/10.110… github.com/theislab/sfaira

Differential effects in #scRNA-seq often are increase of a cell-type - but how to find given issues of compositionality & few samples? Led by @marenbuettner & @schubert_benni, our scCODA model addresses this in a Bayesian framework: biorxiv.org/content/10.110… github.com/theislab/scCODA

Fun end-of-the-year event @ICBmunich & @HelmholtzAI, with >100 people joining in - we had a challenging but successful year, with super-nice people and great science, thanks everyone! Next year f2f again with non-virtual Feuerzangenbowle, pub quiz and maybe even a Santa?

Interesting read @NatMachIntell on issues of fictional AI narratives: "If we understand sci-fi AI too literally [], we get distracted from questions concerning AI that are relevant right now." Just wondered why only movie but no book refs. nature.com/articles/s4225…

Had a short, pre-recorded talk this weekend, but missed (=forgot) the live Q&A on Sat afternoon - was baking cookies with the kids.😮 Cookies turned out well though 🍪😋

Thanks Sten, very kind! Not 100% sure about the ‘fully solved’ part, but what a fun concept you guys gave us modelers 😊

@slinnarsson

”RNA velocity fully solved”: scvelo from @fabian_theis group with @VolkerBergen and @falexwolf - my favorite kind of velocity!

Quoted post

@MDLuecken in group meeting today on metrics to evaluate data integration, ordered from fast to slow execution time - not sure about the axis labels 😂

Really happy to see such a strong paper and for us of course impressive use-case of our transfer learning based integration method scArches, congrats! t.co/CQs4p2Lekh

Quoted post

Impressive new and large scale spatial transcriptomics analysis by John @MarioniLab on mouse organogenesis. SeqFISH with 387 genes, with matched scRNAseq for full imputation. Spatial atlases, they start coming! biorxiv.org/content/10.110…

VIB conference comp session: @ngehlenborg presenting an interactive, web-based visualization tool for spatial single cell genomics, see vitessce.io - soo pretty!

@OliverStegle presenting at VIB single cell genomics conference on his recent impressively large-scale work on modeling differentiating iPS cells. First results at nature.com/articles/s4146…

In SCG, we usually don't have cell phenotype information - FACS-based assays however can image cell morphology. Here we integrate the two to add (rough) gene expression information to existing imaging flow cytometry experiments. Just out in @NAR_Open academic.oup.com/nar/article/48…

VIB single-cell meeting: @oana__ursu from Regev lab @broadinstitute talking about using perturb-seq to study variant impact in cancer - 200 variants, 300k lung cancer cells: biorxiv.org/content/10.110…

Amazing talk by Amos Tanay at VIB Single Cell conf on leveraging single cell epigenomics to study embryonic commitment. He's constructing a temporal atlas, and now following this up with perturbations. Super-interesting computational challenges regarding population modeling!