Lightnews — Scholar-powered news

Stijn De Baerdemacker @cortogantese.bsky.social · 4h

Note to self: if you’re working remote, and the only thing you have is a touchpad and no mouse: middle click is 3(!) fingers! #fedora #firstworldproblems #iwastodayyearsold

Stijn De Baerdemacker @cortogantese.bsky.social · 6d

And on a darker note: it is a failure of society that we need the investment of $B private companies to deliver cornerstones of research and progress #AIsouvereignty #alfafold #omol25

Stijn De Baerdemacker @cortogantese.bsky.social · 6d

Already reposted it on LinkedIn today, (@justinoberman.bsky.social), so might as well throw it in here as well:

“AI is not a replacement of understanding, it’s an amplifier of understanding”

Case in point: you’ll need to understand DFT if you want ML to have an impact in #compchem/#quantchem

2 5

Stijn De Baerdemacker @cortogantese.bsky.social · 7d

Can’t really pinpoint it

Stijn De Baerdemacker @cortogantese.bsky.social · 8d

Why does it look like this fellow is AI- generated?

1 5

Reposted by Stijn De Baerdemacker

Miranda-Quintana group @groupquintana.bsky.social · 13d

The shape of chemical space: www.biorxiv.org/content/10.1...
github.com/mqcomplab/NAMI

The Shape of Chemical Space

The concept of chemical space is critical in cheminformatics, medicinal chemistry, and machine learning applications. Despite this, the high dimensionality of molecular representations greatly complic...

www.biorxiv.org

3 5

Stijn De Baerdemacker @cortogantese.bsky.social · 14d

@chemistryunb.bsky.social

Stijn De Baerdemacker @cortogantese.bsky.social · 14d

Nevertheless, the question got the team engaged in some fun philosophical meanderings, involving undergrad Incé Husain (currently PhD candidate in computational neuroscience @westernu.ca) and QuNB grad Amer El-Samman, among others. Preprint is here: doi.org/10.21203/rs.... (5/5)

Occupancies vs Embeddings: Internal representations of Ab Initio and Graph Probabilistic Models*

Chemical modeling has traditionally been separated into two fairly non-overlapping frameworks: the exact framework of \textit{ab initio} models and the data-driven framework of statistical models. How...

doi.org

1 1

Stijn De Baerdemacker @cortogantese.bsky.social · 14d

and straightforward investigation are not conclusive. Many "hyperparameters" need to be further explored before we can phrase hard claims, such as the rigour of the population analysis (Hirshfeld, etc), the choice of rung on Jacob's ladder, or in/co/equivariance of the ML models. (4/5)

1

Stijn De Baerdemacker @cortogantese.bsky.social · 14d

Furthermore, the predictive power of a simple transfer model from embeddings to occupations proved to be slightly higher than the reverse... Did this answer our question that the ML model picked up on an underlying density governing the (DFT) data? I would say the observations from our simple (3/5)

1

Stijn De Baerdemacker @cortogantese.bsky.social · 14d

in earlier work (doi.org/10.1039/D3DD...) Maybe not as surprising as for embeddings, the (Mulliken & Lowdin) population analyses vectors we looked at also organized themselves geometrically into functional groups, very similar to embedding vectors, however not as crisp! (2/5).

Global geometry of chemical graph neural network representations in terms of chemical moieties

Graph neural nets, such as SchNet, [Schütt et al., J. Chem. Phys., 2018, 148, 241722], and AIMNet, [Zubatyuk et al., Sci. Adv., 2019, 5, 8] provide accurate predictions of chemical quantities without ...

doi.org

1

Stijn De Baerdemacker @cortogantese.bsky.social · 14d

When a Machine Learning model trains on just DFT energies, does it pick up there is an underlying quantity that governs the data: the electron density? We set out to investigate that question, and looked at electron populations in very much the same way we looked at embedding representations (1/5)

1 1 4

Stijn De Baerdemacker @cortogantese.bsky.social · 15d

Chemists: “Everything is Chemistry 🤗”
Chem Nobel: “Join us in celebrating _this_ 🥳”
Chemists: “_this_ is not chemistry 😡”

1 4

Stijn De Baerdemacker @cortogantese.bsky.social · 23d

I have to pass 0 traffic lights, but not all of it was green 🍃🍁😉

1

Stijn De Baerdemacker @cortogantese.bsky.social · 26d

Cool!

Stijn De Baerdemacker @cortogantese.bsky.social · 27d

Then again, would be cool if these structures would _exist_ #everythingyoucanthinkofistrue

1

Reposted by Stijn De Baerdemacker

Nature @nature.com · 28d

We spoke to scientists who left the USA during Trump's first term. Here is their advice

go.nature.com/42oi3pu

These scientists left the US in Trump’s first term: their tips on taking the leap

Nature talks to researchers about why they moved and how they relocated successfully.

go.nature.com

1 21 51

Stijn De Baerdemacker @cortogantese.bsky.social · 28d

Guess ⚽️

2 3

Stijn De Baerdemacker @cortogantese.bsky.social · Sep 8

We had a conversation about (jewish) rituals last night before bedtime

1

Stijn De Baerdemacker @cortogantese.bsky.social · Sep 4

Time, Coffee, Slaters & Staedtler, four of the reasons why I became an academic, I guess… #theochem

Composite picture: handwritten equations on top, coffee cup (bottom left), closeup espresso (bottom middle), inside vars&java cafe (bottom right)

2

Stijn De Baerdemacker @cortogantese.bsky.social · Sep 4

@scipost.bsky.social

Reposted by Stijn De Baerdemacker

Janus Juul Eriksen @januseriksen.bsky.social · Aug 21

The call for the first 2y postdoc position on this new project with a start by early 2026 (Jan/Feb) is now live:

efzu.fa.em2.oraclecloud.com/hcmUI/Candid...

Please share & repost. Deadline is 14 Sept. Copenhagen is a truly wonderful place to live and we offer excellent terms of employment at DTU.

1 2 4

Stijn De Baerdemacker @cortogantese.bsky.social · Sep 3

Couldn’t find her on the website, indeed. Thx for sharing, and have fun! #fomo

Stijn De Baerdemacker @cortogantese.bsky.social · Sep 3

Who’s the artist? 🤖

1

Stijn De Baerdemacker @cortogantese.bsky.social · Sep 2

Lot’s of metaphors in today’s @nytimes.com #spellingbee, but hey! Happy First Day of School! 📚