Florian Huber

link.springer.comCount your bits: fingerprint benchmarking to assess broad chemical space representation - Journal of CheminformaticsQuantifying molecular similarity is a cornerstone of cheminformatics, underpinning applications from virtual screening and nearest-neighbor search to chemical space visualization and the evaluation of...

Our fingerprint analysis work has now been published 🎉 --> link.springer.com/article/10.1... Also check out our small helper library #chemap --> github.com/matchms/chemap (work together with @julianpollmann.bsky.social) #cheminformatics #openscience #Python

Florian Huber reposted

We are offering 4 𝘁𝗲𝗺𝗽𝗼𝗿𝗮𝗿𝘆 𝘁𝘄𝗼-𝘆𝗲𝗮𝗿 𝗽𝗼𝘀𝗶𝘁𝗶𝗼𝗻𝘀 for RSEs to join us on research projects within the context of the @opensciencenl.bsky.social Research Software Sustainability Programme. For more information, please see www.esciencecenter.nl/vacancies/

Florian Huber reposted

@lukaswallraff.bsky.social

The deadline to apply for our #vacancies is fast approaching! If you want to work as a RSE in Life Sciences, or Social Sciences & Humanities, send us your CV, a letter of motivation, and a link to an active GitHub account (or an alternative) before 𝗢𝗰𝘁𝗼𝗯𝗲𝗿 𝟮𝟮𝗻𝗱: www.esciencecenter.nl/vacancies/

Florian Huber reposted

Lukas Wallraff

Reiche Beute @taz.de

Florian Huber reposted

Did you use to work at the eScience Center, and would you like to reconnect with former colleagues and meet new members of the eScience Center team? eScience Center alumni are organising a meetup on November 5th 👉 www.eventbrite.nl/e/escience-a... Help spread the word by sharing this post!

Florian Huber reposted

Tom Gauld

@tomgauld.bsky.social

My latest @newscientist.com cartoon

Florian Huber reposted

tante

@tante.cc

So many philosophy PhDs on staff an nobody told them that you can't be "cruel" to a bunch of statistics. Anthropic bans ‘abusive or cruel behavior’ toward Claude

Florian Huber reposted

bbc.comMargaret Hamilton, whose software helped land Apollo 11 on the moon, dies at 90Pioneering computer scientist led the development of the software for Nasa's Apollo mission and was awarded the Presidential Medal of Freedom.

Pioneering software engineer Margaret Hamilton passed away last week. She coined the term 'software engineering,' helping to establish software development as a legitimate engineering discipline. For this, and for all her other achievements, we are very much in her debt. www.bbc.com/news/article...

Florian Huber reposted

Antonio E. Porreca 🐳

@aeporreca.org

The Association for Human Mathematics (AHM) statement on OpenAI’s October 6 release of mathematical documents just dropped! www.ahmath.org/statements

Florian Huber reposted

@ricardalang.eurosky.social

𝗪𝗲 𝗮𝗿𝗲 𝗵𝗶𝗿𝗶𝗻𝗴! We have vacancies for Research Software Engineers in Life Sciences, and Social Sciences & Humanities. Do you want to work at the forefront of cutting-edge international research? Apply now! www.esciencecenter.nl/vacancies/ The deadline for applications is 22 October 2026.

Florian Huber reposted

Ricarda Lang

Das konservative Bürgertum in diesem Land hat einmal darin versagt, die Gefahr einer rechtsextremen Partei zu erkennen und ihr an die Macht verholfen. Genau jetzt steht jeder in der Union vor der Entscheidung, ob man diesen Fehler wiederholt.

Florian Huber reposted

Ricarda Lang

@ricardalang.eurosky.social

Die Neuauflage des Tankrabatts kostet 2,5 Milliarden. Das ist mehr als alle Einsparungen bei Familien - weniger Elterngeld, Unterhaltsvorschuss und Kindersofortzuschlag - zusammen. Die Kürzungen bei Familien sind kein Sachzwang. Sie sind eine politische Entscheidung.

Die politischen Meldungen der letzten Tage kann ich wirklich bald nur noch in Form von Satire ertragen.

der-postillon.comPeinliche Panne: Merz halbiert versehentlich falsche ParteiDeutschlands größte Tageszeitung der Welt

14 Sept

How to intuitively explore abstract, complex data using interactive #dataviz? --> was great to show people some of our projects at the "Nacht der Wissenschaft" @zdd-hsd.bsky.social

Florian Huber reposted

Der Postillon 📯

@der-postillon.com

6 Sept

Peinliche Panne: Merz halbiert versehentlich falsche Partei www.der-postillon.com/2026/09/cdu-...

7 Sept

Good #Dataviz is very hard to master because even small things can have a big effect on how a visualization is perceived. Today I'd like to thank @taz.de for consistently coloring German election plots correctly. Doesn't make the results any better, but at least it removes the false facade.

taz.deLandtagswahl in Sachsen-Anhalt: Der dicke braune BalkenDie AfD ist klar stärkste Partei, weil sie bei Nichtwählern punktet. BSW erstmals im Landtag. Auffälliges Stimmensplitting bei Linken und Grünen. Alle Grafiken zur Wahl.

7 Sept

Election plot taken from: taz.de/Landtagswahl...

nytimes.comA Drone Killed Three Ukrainians. It Was Guided Entirely by A.I.An attack by what Ukrainian officials said was a Russian drone with an Nvidia chip presages a dystopian future of weaponry untethered to humans.

24 Aug

This comes as no big surprise since it always felt more like a question of *when*, not *if*. I still find it chilling to see this moment where #AI is allowed to decide to kill. www.nytimes.com/2026/08/24/w...

@irisvanrooij.bsky.social

24 Aug

There are great uses for "Edge AI" due to its independence of cloud systems, its energy efficiency and sovereignty. But then there is now this other end of the spectrum. I fear that also the now hyped term "agentic system" could quickly turn sour...

Florian Huber reposted

Jon Worth

@jonworth.eu

24 Aug

Super animation of trains in 🇫🇷🇩🇪🇳🇱🇧🇪🇱🇺🇨🇭 chillchamp1.github.io/github.io/#eu A city like Marseille (France's second or third largest, depending how you define it) has so few trains you can barely see where it is

Florian Huber reposted

Iris van Rooij 💭

19 July

“you not only are the product — yes, they will steal and sell your data — but you are also robbed of learning. When the hype inevitably dies down, as it has done many times before, through AI summers and winters (…), you want to have skills. You want to have a degree that matters.”

Olivia Guest · Ολίβια Γκεστ@olivia.science

also olivia.science/cheating

olivia.scienceWhy does cheating matter?A personal anecdote and an appeal to junior colleagues.

Florian Huber reposted

Martin Holland

@mho.social.heise.de.ap.brid.gy

14 July

Wie schlimm war die #Hitzewelle Ende Juni hierzulande? Ich habe das Gefühl, die Berichterstattung wird dem bislang nicht gerecht. Vielleicht hilft dieses Diagramm? Das sind die Sterbefälle für jeden Tag seit Anfang 2020 – in Rot die Kurve für 2026. https://datawrapper.dwcdn.net/gvL9c/1/

Florian Huber reposted

Jonas Schaible

@beimwort.bsky.social

11 July

Ich stelle mir die Zukunft mit umfassender KI-Nutzung genau so vor, etwa so wie die Deutsche Bahn heute. Im Großen und Ganzen funktioniert alles, meistens jedenfalls, aber schlechter, als es müsste. Immer wieder ist man der Verzweiflung nahe, oder einfach nur fürchterlich genervt. Neuer Newsletter:

open.substack.comDas hier wäre fast ein Anti-KI-Manifest geworden, aber ich schreibe keine ManifesteDie entscheidende Frage lautet: Rechtfertig der Nutzen den Preis?

Here's v1.0 of my data science script 🚀 Last week of my introduction to #datascience course this semester. I used this as another chance to reiterate, improve, expand my teaching materials. The entire course material --> florian-huber.github.io/data_science... #oer #openscience #python #teaching

github.comGitHub - florian-huber/data_science_course: Hands-on Introduction to Data Science with PythonHands-on Introduction to Data Science with Python. Contribute to florian-huber/data_science_course development by creating an account on GitHub.

Finally, here the GitHub link to all source code, notebooks, images etc. github.com/florian-hube...

btw: In one of the last sessions, we trained a small Word2Vec model on 40k restaurant reviews. To illustrate what word embeddings mean, we looked at them through an interactive UMAP plot (see screencast). I know, we live in transformer/LLM age. But I believe you should understand the foundations.

View full thread

6/ For several tasks, including chemical space visualization (using UMAP), bioactivity, or chemical class prediction, MAP4 and RDKit are also very strong candidates. MAP4 excels in particular when it comes to specificity, but at the same time requires the most care when handling properly (see (2)).

7/ Finally, we built a helper Python library: #chemap for handling all the different fingerprint types and variants (including unfolded and frequency-folded), and also allows fast fingerprint and score computations. Contributions/Feedback are very welcome! --> github.com/matchms/chemap

View full thread

4/ (2) High-bit-number fingerprints (e.g., RDKit, MAP4) tend to strongly favor larger over smaller compounds and produce arbitrary high-similarity predictions. This can be fixed by simply using unfolded variants! We also proposed and explored "frequency-folded" fingerprints for ML tasks.

5/ (3) There is no clear "winner", so no fingerprint that does better on all or our benchmarking tasks. When in doubt, however, Morgan and FCFP count fingerprints, ideally with a larger radius, are good starting points. They also correlate well with graph-based MCES similarities.

2/ Some key findings: (1) Use count fingerprints! (nearly always better than binary) (2) Consider alternatives to classical folding. (3) No clear "winner", but larger radius Morgan or FCFP fingerprints are often good first bets.

3/ (1) Counting substructure or bit occurrences drastically increases specificity. For binary fingerprints, we typically observed a high fraction of identical fingerprints for notably differing compounds. Also, chemical class prediction and neighborhood consistency improve when using counts.

19 May

Another round of edits on my "Hands-on Introduction to Data Science" book. Finally managed to add a bit more explanatory text and figures on UMAP in the dimensionality reduction chapter. --> florian-huber.github.io/data_science... #OpenScience #teaching #python #datascience #oer

1 Apr

Working on another iteration of my #DataScience introductory course; Revising/expanding materials and new figures and illustrations. Here is a new sketch to display the rich data science skills landscape (not meant to be exhaustive). --> The book: florian-huber.github.io/data_science...

github.comGitHub - matchms/ms2deepscore: Deep learning similarity measure for comparing MS/MS spectra with respect to their chemical similarityDeep learning similarity measure for comparing MS/MS spectra with respect to their chemical similarity - matchms/ms2deepscore

31 Mar

New MS2DeepScore release (2.9.0) 🚀 --> github.com/matchms/ms2d... The main change is the ability to use count and log-count rdkit fingerprints, as well as unfolded fingerprints for training. #opensource #massspec #Python

Florian Huber reposted

Jonathan Stray

@jonathanstray.com

27 Mar

Could social media make us less polarized instead of more? We tested 5 algorithms on 3 platforms with 10,000 people for 6 months during the 2024 election, and found that the answer is yes. 🧵

Florian Huber reposted

Greg Landrum

@greglandrum.bsky.social

27 Mar

Free registration for the 2026 #RDKit UGM (both in-person and online attendance) is now open: www.eventbrite.com/e/1985889262...

eventbrite.com15th RDKit UGM 20262026 RDKit User Group Meeting (in person and online)

Florian Huber reposted

404 Media

@404media.co

26 Mar

NEW: Wikipedia has banned AI-generated content.

404media.coWikipedia Bans AI-Generated Content“In recent months, more and more administrative reports centered on LLM-related issues, and editors were being overwhelmed.”

nytimes.comThe Billionaire Funding France’s Far Right

22 Mar

Yet another example of why too much wealth means too much power. Remaining billionaire should fiscally not be possible. www.nytimes.com/2026/03/22/w... #taxtherich

Florian Huber reposted

Ingwar Perowanowitsch

@ingwarpero.bsky.social

21 Mar

Heute war der Startschuss eines tollen Projekts. Es heißt „Abpflastern“ und kam 2025 aus den Niederlanden nach Deutschland. Es ist ein Wettbewerb zwischen Städten und Gemeinden. Wer vom 21. März bis 31. Oktober am meisten Flächen entsiegelt & begrünt, gewinnt. Mitmachen können Privatpersonen...(1/2)

Florian Huber reposted

Daniel Petras

@daniel-petras.bsky.social

17 Mar

Our new paper on the presence of xenobiotics in marine dissolved organic matter just come out. Thanks to Jarmo Kalinski and our awesome collaborators, we were able to reanalyze more than 20 public LC-MS/MS datasets from seawater and ask how many anthropogenic compounds we can detect. rdcu.be/e8q6C

Florian Huber reposted

Justin J.J. van der Hooft

@jjjvanderhooft.bsky.social

17 Mar

www.linkedin.com/posts/jjjvan...

linkedin.com#metabolomics #massspectrometry #networking #embedding #proudpi #compmetabolomics #bioinformatics | Justin J.J. van der HooftMany congrats 👏 🎉 🙌 to Niek de Jonge for initializing the concept of MS2DeepScore 2.0 😎 Please read Niek's post for all the innovations and updates we made - most importantly, this brings the combine...

nature.comCross ionization mode chemical similarity prediction between tandem mass spectra in metabolomics - Nature CommunicationsMass spectrometry is a cornerstone of untargeted metabolomics, but comparisons across ionization modes have remained a substantial challenge due to the distinct fragmentation patterns produced by each...

17 Mar

MS2DeepScore 2.0 is finally published 🚀 --> www.nature.com/articles/s41... This was a great journey with Niek de Jonge and a great team of collaborators! See more on LinkedIn: www.linkedin.com/posts/f-hube... #massspec #cheminformatics #ML #opensource #openscience #python

14 Mar

Bit ironic that we cannot name the new generation (of humans) GenAI because even that term is already taken by AI.

Florian Huber reposted

Robin Berjon

@robin.berjon.com

1 Mar

Replying to @robin.berjon.com

And @vestager.bsky.social takes the stage at #Rebuild. She's talking about digital infrastructure: "That's not a tech problem. That's a democratic problem." It's amazing to finally hear this said by someone with voice.

27 Feb

Just updated our preprint on benchmarking molecular fingerprints! --> www.biorxiv.org/content/10.1... Some key points - Count fingerprints should be the default (not binary!) - Unfolded fingerprints are often worth it. - Larger radius Morgan or FCFP fingerprints are good first bets.