iiinova — where your digital self keeps evolving

Gregor Sturm @grst.bsky.social · 05/08/2026

It turns out @nextflow.io is useful outside bioinformatics. Also kudos to the karttapullautin devs for their awesome tool, which the pipeline relies on for the tile rendering step. github.com/karttapullau...

github.com

GitHub - karttapullautin/karttapullautin: Source code of the rust implementation of Karttapullautin

Source code of the rust implementation of Karttapullautin - karttapullautin/karttapullautin

010

Gregor Sturm @grst.bsky.social · 05/08/2026

Using a Nextflow pipeline, I processed 15 TB of LIDAR data + OpenStreetMap shapes into a 180GB tile pyramid.

100

Gregor Sturm @grst.bsky.social · 05/08/2026

This time not bioinformatics, but geoinformatics: I created MapAnt Bayern, an automatically generated orienteering map of Bavaria: mapant.orienteering-allgaeu.de

131

Gregor Sturm @grst.bsky.social · 27/07/2026

@daweonline.bsky.social does that mean that schist becomes usable on larger datasets now?

220

Gregor Sturm @grst.bsky.social · 29/05/2026

Also thanks to @francescafinotello.bsky.social for supporting this and @thehumanborch.bsky.social for providing early feedback!

020

Gregor Sturm @grst.bsky.social · 29/05/2026

This started as a collaboration with biocypher (@slobentanzer.bsky.social) at a @scverse.bsky.social hackathon, was then advanced by @valeriiadragan.bsky.social as a master thesis project and finally pushed over the finish line by Raphael De Gottardi

222

Gregor Sturm @grst.bsky.social · 29/05/2026

Happy to announce the release of IggyTop, a metadatabase for immune receptor–epitope interactions. It's integrated with scirpy v0.24 so you can readily use it to annotate your single-cell TCR datasets. iggytop.readthedocs.io/en/latest/

198

Gregor Sturm @grst.bsky.social · 17/04/2026

sorry, I meant min, not max

000

Gregor Sturm @grst.bsky.social · 17/04/2026

Thanks for the response! The thing is that the original TCRdist uses max(4, 4-score) so it caps scores at 4. With TCRblosum that would be max(2, 2-score) which would destroy all the strong signal that you have in e.g. the C residue. So you'd suggest to use 2-score without the max instead?

200

Gregor Sturm @grst.bsky.social · 14/04/2026

Hi @pmeysman.bsky.social‬, we are trying to integrate tcrBLOSUM into scirpy, our library for scTCRseq analysis. Specifically, we want to adapt the TCRdist algorithm to use tcrBLOSUM substitution values. How did you turn the substitution matrix into a distance matrix in your paper?

200

Gregor Sturm @grst.bsky.social · 22/08/2025

There's another scverse conference this year and it will be amazing! Register now: www.eventbrite.com/e/scverse-co...

021

Gregor Sturm @grst.bsky.social · 19/08/2025

AFAIK, these differences are minor, numeric differences. I would consider them equivalent.

130

Gregor Sturm @grst.bsky.social · 13/08/2025

Our benchmark + guidelines for atlas-level differential gene expression of single cells is online: academic.oup.com/bib/article/... Bottom line: Use pseudobulk + DESeq2 in simple and pseudobulk + DREAM in more complex settings. Collab w/ @leonhafner.bsky.social @itisalist.bsky.social

1156

Gregor Sturm @grst.bsky.social · 05/08/2025

000

Reposted by Gregor Sturm

scverse @scverse.bsky.social · 12/05/2025

📣 Mark your calendars! The 2025 edition of the scverse conference will take place on 17-19 November at Stanford University (US) scverse.org/conference20... Call for abstracts and registrations coming soon!

scverse.org

scverse conference 2025

1129

Gregor Sturm @grst.bsky.social · 02/04/2025

Just released a new version of the @scverse.bsky.social cookiecutter template: github.com/scverse/cook... Some highlights: 🔃 improved template sync (merge conflicts now show up as such) 🚀 use hatch as project manager 🔧 lots of fixes and documentation updates

github.com

Release v0.5.0 · scverse/cookiecutter-scverse

New template sync We re-implemented template sync from scratch instead on relying on cruft. This allows us to create real merge conflicts that show up as such on GitHub instead of .rej files. Gene...

042

Gregor Sturm @grst.bsky.social · 14/03/2025

Nice post! How did you generate the doi-link for a blog post?

100

Reposted by Gregor Sturm

Wolfgang Huber @wkhuber.bsky.social · 04/03/2025

Blog post by @const-ae.bsky.social with a simple explanation of the manifold regression algorithm & code that underlies our paper “Analysis of multi-condition single-cell data with latent embedding multivariate regression” (doi.org/10.1002/eji....). const-ae.name/post/2025-01...

const-ae.name

LEMUR simplified | const-ae

A simplified implementation of the LEMUR algorithm.

1255

Reposted by Gregor Sturm

scverse @scverse.bsky.social · 14/02/2025

🎉 Scanpy 1.11.0 is out! 🎉 just after reaching 2000 stars on GitHub! - sc.pp.sample replaces subsample with many new features - Sparse Dask support pca - session-info2 package for more reproducible notebooks See the release notes:

buff.ly

Release notes

Version 1.11: 1.11.0 2025-02-14: Release candidates: rc2 2025-01-24, rc1 2024-12-20. Features: rc1 sample() supports both upsampling and downsampling of observations and variables. subsample() is n...

14919

Reposted by Gregor Sturm

Edmund Miller @edmundmiller.dev · 09/02/2025

Been looking forward to this talk since @alexpeltzer.bsky.social told me about DSO in October!

041

Reposted by Gregor Sturm

Gregor Sturm @grst.bsky.social · 05/02/2025

I'd like to share DSO, a command line helper to build reproducible data science projects with ease. It is an opinionated way to organize data science projects, built around data version control (DVC). github.com/Boehringer-I...

github.com

GitHub - Boehringer-Ingelheim/dso: Data Science Operations (dso) command line tool

Data Science Operations (dso) command line tool. Contribute to Boehringer-Ingelheim/dso development by creating an account on GitHub.

1124

Gregor Sturm @grst.bsky.social · 05/02/2025

We try to avoid that by using this with preprocessed data only. All the heavy lifting is done with nextflow pipelines before. Datasets up to tens of GBs have worked well so far.

020

Gregor Sturm @grst.bsky.social · 05/02/2025

Finally, many thanks to my colleagues @alexpeltzer.bsky.social, Daniel Schreyer and Tom Schwarzl for testing, adopting, and contributing to DSO.

110

Gregor Sturm @grst.bsky.social · 05/02/2025

If you want to learn more, I'll be presenting this at a @nf-co.re bytesize talk: nf-co.re/events/2025/...

nf-co.re

Bytesize: data science operations (DSO)

Gregor Sturm, Boehringer Ingelheim

152

Gregor Sturm @grst.bsky.social · 05/02/2025

We built this at @boehringerglobal.bsky.social to meet the quality standards required for biomarker analysis in clinical trials. But I think this is useful for any kind of data analysis project.

110

Gregor Sturm @grst.bsky.social · 05/02/2025

One of my favorite features: automated watermarking of all plots in a quarto report. Nobody gonna publish my plots anymore before I think they are ready.

An exemplary PCA plot with a "preliminary" watermark.

130

Gregor Sturm @grst.bsky.social · 05/02/2025

It brings together the best tools: - git, for code versioning - dvc, for data versioning and tracking inputs and outputs - jinja2, for templates - uv, for Python dep mgmt - quarto, for authoring reports - hiyapyco, for hierarchical YAML config - pre-commit, for linting

110

Gregor Sturm @grst.bsky.social · 05/02/2025

github.com

GitHub - Boehringer-Ingelheim/dso: Data Science Operations (dso) command line tool

Data Science Operations (dso) command line tool. Contribute to Boehringer-Ingelheim/dso development by creating an account on GitHub.

1124

Reposted by Gregor Sturm

Stefano Mangiola @stemang.bsky.social · 22/01/2025

We (Chen Zhan!) just launched #sccomp for #Python! Testing for differences in cell-type proportion in #singlecell #spatial data? #sccomp is a mixed-effect Bayesian model - Use sum-constrained BetaBinomial distribution - Outliers detect. - Remove unwanted effects github.com/MangiolaLabo...

1123

Gregor Sturm @grst.bsky.social · 19/01/2025

(2) Finding the mistake, tracing it back to its origin, and fixing it was only possible because the data and scripts for building the atlas are publicly available and fully reproducible. github.com/icbi-lab/luca

github.com

GitHub - icbi-lab/luca: Single-cell Lung Cancer Atlas with 1.2M cells

Single-cell Lung Cancer Atlas with 1.2M cells. Contribute to icbi-lab/luca development by creating an account on GitHub.

020

Gregor Sturm @grst.bsky.social · 19/01/2025

(1) Maintaining a data resource is very much like maintaining software. It is never "done" but constantly improving.

120

Gregor Sturm @grst.bsky.social · 19/01/2025

Two years after publication of our single-cell lung cancer atlas, a user found a mistake in the annotation of the EGFR-status of some patients. We fixed the issue and the atlas is now updated on cell-x-gene: cellxgene.cziscience.com/collections/... What are the takeaways from that? (1/3)

cellxgene.cziscience.com

Cellxgene Data Portal

Find, download, and visually explore curated and standardized single cell datasets.

273

Reposted by Gregor Sturm

Constantin Ahlmann-Eltze @const-ae.bsky.social · 03/01/2025

After 4y in the making, I am super excited that my main PhD project is published 🎉🥳🎉🎉🥳 www.nature.com/articles/s41... LEMUR is a tool to analyze multi-condition single-cell data and model differential expression as a continuous function of the cell-state space. Some highlights⬇️

Overview of the LEMUR steps: (1) subspace alignment, (2) differential expression, (3) DE neighborhoods, (4) pseudobulking.

815435

Gregor Sturm @grst.bsky.social · 02/01/2025

The big issue here in Germany is that we pay ~20 ct/kWh in fixed network fees and tax. Really limits how much you can save.

100

Gregor Sturm @grst.bsky.social · 02/01/2025

While definitely interesting, dynamic tariffs have a much higher cost-saving potential for devices that consume a lot of energy and are easy to regulate automatically. Such as a heat pump or electric car - of which we have neither, for now.

100

Gregor Sturm @grst.bsky.social · 02/01/2025

There's a certain price risk, though. In the energy crisis 2021-2022 prices increased significantly. However prices in 2023 were down to normal, while many fixed price tariffs increased their rates.

100

Gregor Sturm @grst.bsky.social · 02/01/2025

Dynamic electricity tariffs are an incentive to use energy when it's abundant and emits little CO2. But are they also cheaper? ✅ For 2024, we would have saved 10-13% compared to our current fixed price tariff. Without any optimization. Full post (in German): grst.github.io/dynamischer-...

Monthly comparison of fixed price tariff (AÜW) with dynamic tariff (tado).

120

Gregor Sturm @grst.bsky.social · 30/12/2024

Modern tar detects the compression automatically when reading from a file. So `tar xvf` covers most of the cases already.

040

Reposted by Gregor Sturm

khrovatin.bsky.social @khrovatin.bsky.social · 16/12/2024

To bring to light data science topics that usually don’t make it into publications I started a blog on this topic: hrovatin.github.io By interviewing different researchers, I plan to find out what is going on in the community.

hrovatin.github.io

Karin Hrovatin

Data science blog on topics that don’t get published.

1106

Reposted by Gregor Sturm

Gregor Sturm @grst.bsky.social · 07/12/2024

Formulaic is the go-to way to specify design formulas in Python, e.g. ~treatment + timepoint. To compare sth, one needs to specify a contrast, e.g "on treatment vs baseline". To make this easier, we developed "formulaic-contrasts": formulaic-contrasts.readthedocs.io/en/latest/

# Define model with interaction term
model = MyModel(data, "~ treatment * timepoint")

# compare timepoints
contrast = model.cond(timepoint="on_treatment") - model.cond(timepoint="baseline")

# compare timepoints within drugA only
contrast = (
model.cond(treatment="drugA", timepoint="on_treatment") -
model.cond(treatment="drugA", timepoint="baseline")
)

# compare interaction of timepoint with treatment
# (= difference of changes between both treatments)
contrast = (
mod.cond(treatment="drugB", timepoint="on_treatment")
- mod.cond(treatment="drugB", timepoint="baseline")
) - (
mod.cond(treatment="drugA", timepoint="on_treatment")
- mod.cond(treatment="drugA", timepoint="baseline")
)

1186

Gregor Sturm @grst.bsky.social · 07/12/2024

Formulaic-contrasts was originally conceived at the @scverse.bsky.social hackathon in Cambridge based on an idea by @const-ae.bsky.social. It is currently (being) integrated into #pyDESeq2, #pertpy and #pyLemur.

020

Gregor Sturm @grst.bsky.social · 07/12/2024

1186