Posts on X

If you are interested in analyzing single cell genomics data, please consider coming to Munich in late March for our #SCOG workshop.

@singlecellomics

Register for our upcoming workshop on computational single cell genomics, sponsored by SCOG @singlecellomics and chaired by @fabian_theis . Only few places left. singlecell.de/index.php/work…

Quoted post

Thx, lovely! Next step: branching analysis on morphometric readouts.

@DrAnneCarpenter

@ZanotelliVRT Nice! @fabian_theis look: Some use of pseudotime on images!

Quoted post

He (?) has your eyes, Marius 👀

@CompHealthMuc

1st @ICBmunich snowman championship started today, the biggest snowman wins. Hard to beat! #WinterWonderland #ICBchallengeofthemonth

Quoted post

Awesome bike commute to @HelmholtzMucEn today - beautiful winter!

I think many universities & research institutes are starting to get this - eg we have a tt junior group system @HelmholtzMucEn and career professor track @TU_Muenchen. Germany needs to solve still too heavy & non-digital bureaucracy though.

@dagarfield

@OliverStegle @fabian_theis True, but I would argue that the European system of new academic jobs every 3-5 years for lack of a proper tenure track system is just as damaging. European society is better for families (and perhaps better period!). But the academic system has its own work to do.

Quoted post

Agreed, this is actually often the hardest part. Two-body or with kids n-body problems are tough to solve.

@KrishnaswamyLab

@fabian_theis Sounds good! Now you'll just have to convince the rest of my family to move.

Quoted post

Why not come to Europe, where we have public health insurance, paid parental leave and often cheap daycare? (Higher taxes though)

Quoted post

I agree - using scRNA-seq time series one can deconvolute this. Methods for this are eg Waddington-OT from @eric_lander lab or (shameless ad) pseudodynamics from us biorxiv.org/content/10.110…

@slinnarsson

@fabfrohlich @fabian_theis @KharchenkoLab No reason for a high cell density. There might be very distinct cell types present in only a few copies (e.g. oocytes, or certain hypothalamic neurons). Don’t conflate cell density with local stability. Density is history-dependent.

Quoted post

Not necessarily, depends on apoptosis and sampling - we discussed this in Laleh Haghverdi’s diffusion pseudotime paper).

Quoted post

Not sure how we define an attractor without any dynamical system on the manifold. Having a vector field such as RNA velocity from @slinnarsson and @KharchenkoLab of course allows to do this.

@mkleyman_cb

@slinnarsson @KharchenkoLab @fabian_theis What about defining cell types as attractor states in cell state space/manifolds?

Quoted post

Good summary of the discussion - also liked Nicole Rusk’s comment that editors encourage presentation of limitations of presented tools. We have to make assumptions for any algorithm (-> no-free-lunch theorem bit.ly/1QomNkp), so let’s talk about it!

@satijalab

Blog post by Nicole Rusk (@naturemethods ) summarizing a keystone panel discussion w/myself, @fabian_theis @dana_peer, and Allon Klein : “What is the next big thing in single cell data analysis, and what is holding us back?” protocolsmethods.springernature.com/channels/1913-…

Quoted post

Really interesting to combine molecular and morphological readouts (how?) - but not sure that tSNEs would tell you much about cell types besides visualizing clusters (from higher dim analysis). t.co/0zHjrfNCli

Quoted post

Awesome blog post - @lpachter games the system and raises download metric of a biorxiv paper 100fold. While I thought they would at least check for unique IPs, I of course agree that any such metric can be fooled - combat the ‘proliferation of technologies of narcissism’!

@lpachter

@davisjmcc @fabian_theis @naturemethods Did you actually do it? If not you can download one from here: liorpachter.wordpress.com/2019/01/29/tec…

Quoted post

Great to see some Starcraft fans here :)

@MDLuecken

@GCLinderman @fabian_theis @DeepMindAI This is exactly what I thought when I saw game 2 against MaNa... and I have no knowledge of Starcraft 2. The micro was incredible though... not human replicable.

Quoted post

Very interesting point and true that the micro was superhuman. But still the AI did do some quite good macro decisions as well, which cannot be explained by high APM I guess.

@GCLinderman

@fabian_theis @DeepMindAI Very impressive work by DeepMind, no question. But here is an interesting analysis that suggests it might not be as fair of a test as it seemed. blog.usejournal.com/an-analysis-on…

Quoted post

StarCraft - an old-school real-time strategy game (sank many hours into that) - now joins Chess and Go and gets conquered by our computer overlords. Really worthwhile watching and loads of fun. Congrats, @DeepMindAI. youtube.com/watch?v=cUTMhm…

Great short read on algorithmic bias independent on human bias in labeling and cost function design: ‘Well intentioned "fairness" regulations prohibiting decision makers form taking sensitive attributes into account can actually make things less fair and less accurate’

@Aaroth

A short blog post about how unfairness can creep into algorithmic predictions, without any bias being baked in by the algorithm designer or human data labelers: aaronsadventures.blogspot.com/2019/01/discus… 1/3

Quoted post

We like diffusion pseudo time for that (an actual distance, determined in the kNN graph)

@s_anders_m

@rFarouni @mikelove So, we should add to sleepwalk some way to also compare different feature-space distances. What best to use for an estimated on-manifold geodesic distance?

Quoted post

Data sharing: ‘Across the entire biomedical literature, in 2015 to 2017, about one-fifth of published articles shared raw data, a major increase over previous years’

@mason_lab

Free the data! The time has come to do it faster, we argue in our new paper science.sciencemag.org/content/363/64…

Quoted post

Really interesting! But can I now use this statistics to convince @naturemethods to take my paper once it has reached 2000 downloads on biorxiv?

@louisaflintoft

Interesting insights from bioRxiv’s first 30,000 preprints, incl correlation between downloads and IF of eventual journal destination nature.com/articles/d4158…

Quoted post

@MatthiasTschop summarizing 2018's science @HelmholtzMucEn's townhall meeting. Great year, in particular with major focus on young scientists: two new grad schools in diabetes (helmholtzresearchschool-diabetes.org) and data science (mu-ds.de).

Our paper on finding efficient latent spaces in scRNA-seq using #NeuralNetworks has just come out. Adapting error function to negative binomial helped a lot. Thanks @gokcen and @lkmklsmn! nature.com/articles/s4146…

Interesting! - the model looks like a nonnegative matrix factorization in gene expression space. Am I missing s.t., in particular any regulatory explanation possible for such continua arising, say instead of bistable/multi stable states at the vertices? Just initial conditions?

@UriAlonWeizmann

Our new paper: division of labor within a single cell type can lead to a continuum of gene-expression inside a polyhedron. The vertices of the polyhedron reveal what the cell’s different tasks are. Specialist cells have specific positions in the tissue. authors.elsevier.com/a/1YRu48YyDfV0…

Quoted post

So is it mostly a political challenge? Or an economic one? Really wonder where scientists can contribute (going political or startup? :)

@ErwinBottinger

„There is no system and we need an entirely new ecosystem“ and „there are huge vested interests opposing any change“ Calum MacRae states the most critical challenges for digital medicine #KSdighealth @HPI_DE @PersonalizedMed @IcahnMountSinai @BronfmanInst @jdudley

Quoted post

Really interesting read - particularly liked @ZavolanLab‘s perspective on the importance of data integration. She questions mechanistic models due to missing parameters and calls for automated methods to process large data across organs, time and perturbations.

@GenomeBiology

We asked members of our editorial board what they thought was going to happen next in genomics. This is what they said: genomebiology.biomedcentral.com/articles/10.11…

Quoted post

And Fortran

@_odsc

An argument for why every data scientist should know SQL. #DataScience #SQL hubs.ly/H0gcyC_0

Quoted post

Exciting set of speakers for #ICANN2019, the 28th International Conference on Artificial Neural Networks! Looking forward to seeing you in Munich in September: icann2019.org #AI #MachineLearning @EuNeuralNetSoc

Surely you have heard that data science is the 'sexiest job of the 21st century' - may be slightly overhyped, after all. Great explanation by @Jonny_CBB why so many data scientists are leaving their jobs: bit.ly/2E7rvji

Great talk at #KSsinglecell by Thomas Höfer @DKFZ, showing how to integrate barcodes into models of population dynamics. Super-interesting to combine with molecular readouts from scRNAseq, since many effects are seen as shifts of cell numbers per cell type cluster.

Rick Horwitz now talking at #KSsinglecell on large team-based effort of the @AllenInstitute for Cell Science. Last year they went through a "computational tool building phase - next year comes the fun". Hey, we comp people have fun, too. You should hear our jokes.