Lightnews — Scholar-powered news

Simon Mathis

@simonmathis.bsky.social

910 followers 580 following 31 posts

PhD student at Uni of Cambridge, UK 🔬 | AI for protein design & engineering 🧬 | biotech & environmental applications 🌱 | enzymes 🏗️ 🇦🇹🇨🇭

Posts Media Videos Starter Packs

Pinned

Simon Mathis @simonmathis.bsky.social · Nov 20

🧪 For people interested in AI & enzymes (enzyme engineering, design, discovery, ...), I'm assembling a starter pack for us.

DM if you'd like to be included!

go.bsky.app/MhfaQBh

6 8 36

Reposted by Simon Mathis

Nate Corley @ncorley.bsky.social · Aug 15

(1/7)
Training biomolecular foundation models shouldn't be so hard. And open-source structure prediction is important. So today we're releasing two software packages: AtomWorks and RosettaFold3 (RF3)

[https://www.biorxiv.org/content/10.1101/2025.08.14.670328v2](www.biorxiv.org/content/10.1...)

Accelerating Biomolecular Modeling with AtomWorks and RF3

Deep learning methods trained on protein structure databases have revolutionized biomolecular structure prediction, but developing and training new models remains a considerable challenge. To facilita...

www.biorxiv.org

2 28 67

Reposted by Simon Mathis

Chaitanya K. Joshi @chaitjo.bsky.social · Aug 15

RosettaFold 3 is here! 🧬🚀

AtomWorks (the foundational data pipeline powering it) is perhaps the really most exciting part of this release!

Congratulations @simonmathis.bsky.social and team!!! ❤️

bioRxiv preprint: www.biorxiv.org/content/10.1...

19 52

Reposted by Simon Mathis

Tomáš Pluskal @pluskal-lab.org · May 24

This paper represents a great effort by @roman-bushuiev.bsky.social and his brother @anton-bushuiev.bsky.social. The DreaMS foundation model for mass spectra of small molecules now opens lots of avenues for possible downstream applications. It might be a game changer for computational metabolomics.

Nature Biotechnology @natbiotech.nature.com · May 23

Self-supervised learning of molecular representations from millions of tandem mass spectra using DreaMS - @pluskal-lab.org @iocbprague.bsky.social go.nature.com/4k1n5iC

Self-supervised learning of molecular representations from millions of tandem mass spectra using DreaMS - Nature Biotechnology

A transformer model is used to construct the DreaMS Atlas—a molecular network of 201 million MS/MS spectra.

go.nature.com

20 52

Simon Mathis @simonmathis.bsky.social · May 10

Very nice, thoughtful post - I really enjoyed the read @pascalnotin.bsky.social

Pascal Notin @pascalnotin.bsky.social · May 8

Even simple methods leveraging these 2 modalities significantly outperform billion-parameter sequence-only models. So, what's next? Better retrieval, advanced multimodal approaches, & alignment. Read more: pascalnotin.substack.com/p/have-we-hi... #BioTech #AI #pLMs

Have We Hit the Scaling Wall for Protein Language Models?

Beyond Scaling: What Truly Works in Protein Fitness Prediction

pascalnotin.substack.com

Reposted by Simon Mathis

briantrippe.bsky.social @briantrippe.bsky.social · Feb 19

🔥 Benchmark Alert! MotifBench sets a new standard for evaluating protein design methods in motif scaffolding.
Why does this matter? Reproducibility & fair comparison have been lacking—until now.
Paper: arxiv.org/abs/2502.12479 | Repo: github.com/blt2114/Moti...
A thread ⬇️

1 17 40

Reposted by Simon Mathis

Adrian Bunzel @bunzela.bsky.social · Feb 12

🚨Preprint Alert!🚨

#ProteinDesign is advancing rapidly—wouldn't it be great to seamlessly combine design tools to achieve more than what each can do alone?🤔

Here, we introduce AI.zymes: A modular platform for evolutionary #EnzymeDesign.♻️🖥️

biorxiv.org/content/10.1...
1/🧵

1 7 30

Simon Mathis @simonmathis.bsky.social · Dec 23

Amazing, congratulations @machine.learning.bio

Reposted by Simon Mathis

Gabriele Corso @gcorso.bsky.social · Dec 8

You may have seen a recent pre-print [1] from Jain et al. with strongly worded claims against the experimental results in our DiffDock paper [2]. We initially declined to respond as we saw that this preprint contained falsehoods, misleading comparisons, seemingly deliberate omissions, ...1/n

2 11 30

Reposted by Simon Mathis

owl @owlposting1.bsky.social · Dec 4

Can AI improve the current state of molecular simulation?

www.owlposting.com/p/can-ai-imp...

in my first podcast, I spend 2 hours interviewing Corin Wagen and Ari Wagen, two brothers who are building the next generation of molecular simulation for drug discovery and material science

Can AI improve the current state of molecular simulation? (Corin & Ari Wagen, Ep #1)

2.1 hours listening time

www.owlposting.com

1 14 56

Simon Mathis @simonmathis.bsky.social · Dec 4

Thank you for the feedback! That’s great to hear 🙌

Simon Mathis @simonmathis.bsky.social · Dec 2

(2/2) ... say a diffusion trajectory then it's doing something you cannot achieve by rsyncing folders (or only *very* cumbersomely). I mostly use it to debug & sanity check my code whilst developing for example

Simon Mathis @simonmathis.bsky.social · Dec 2

(1/2) Good question! If you only use it to look at static pymol files that you saved out then yes, it's an alternative to rsyncing your hpc folder. If you use it to for example visualize in-RAM objects during code execution / debugging, or if you use it to manually dock something midway through ...

Reposted by Simon Mathis

Philippe Schwaller @pschwllr.bsky.social · Dec 2

We are hiring (resharing appreciated)!

Given recent successful grant applications (I got my SNSF Starting Grant 🚀), we are extending the LIAC team with multiple openings (PhD/postdoc) for 2025.

Apply now (deadline: December 20th) by filling in this form: forms.fillout.com/t/eq5ADAw3kkus.
#ChemSky

6 71 100

Simon Mathis @simonmathis.bsky.social · Nov 30

Thank you Greg! I actually found out about this functionality by using your fantastic RDKit package and very much based it on your RPC implementation there ( 😉 see the 3. Credits section). I mainly added functionality to send back and forth application states and a wrapper command for ease of use

Simon Mathis @simonmathis.bsky.social · Nov 30

Haha great point ^^ your milage varies but I've found Claude sonnet 3.5 via the cursor.com integration to work reasonably well -- the copilots then pick up patterns if there's already some sensible commands within the context

Cursor

Built to make you extraordinarily productive, Cursor is the best way to code with AI.

www.cursor.com

1 1

Simon Mathis @simonmathis.bsky.social · Nov 29

For some more guidance on how to use this, Martin Buttenschön wrote a nice blogpost: www.blopig.com/blog/2024/11...

1 4 18

Reposted by Simon Mathis

Roman Bushuiev @roman-bushuiev.bsky.social · Nov 27

Check out our MassSpecGym dataset on @polarishq.bsky.social. 🤩

Cas @caswognum.nl · Nov 27

Wow! 🤩

This may be the most carefully documented dataset on @polarishq.bsky.social. Great work by @roman-bushuiev.bsky.social!

Do I have any #MassSpec researchers in my 🦋 network yet? I would love to hear what you think! github.com/polaris-hub/...

#mass #spectrometry #dataset #benchmark

massspecgym

roman-bushuiev

polarishub.io

3 10

Reposted by Simon Mathis

Diego del Alamo @delalamo.xyz · Nov 27

Conformational dynamics smoothen the fitness landscapes of enzymes. Who knew? 🧪🧶

Another important biophysical factor underlying how fitness landscapes are traversed during evolution is conformational sampling, i.e. the ability of proteins to adopt multiple conformational substates.48,49 Different conformational substates could have different activities and thus contribute differently to the net fitness of the protein: conformation “A” could represent 5% of the populated states, but catalyze a new reaction that the other conformations cannot. Evolution can then act to smoothly “tune” the function of the protein through remote mutations that shift its conformational equilibrium by relatively stabilizing conformation “A”. This results in a smoother landscape because there are generally many more pathways by which a pre-existing conformational equilibrium between two conformations can be affected by remote mutations than the number of solutions that are available through complete remodeling of an active site. For example, the laboratory evolution of phosphotriesterase into an into arylesterase was tracked with protein crystallography, which revealed that almost all the mutations were remote from the active site and shifted the conformational equilibrium to favor pre-existing states that were beneficial to the new catalytic activity (50). The same process was observed in the evolution of a computationally designed Kemp eliminase, wherein gradual changes in the conformational sampling between inactive and active states resulting in a remarkably smooth evolutionary transition (51). These examples of evolutionary trajectories generated by laboratory evolution demonstrate how conformational flexibility can produce smooth fitness landscapes, and help explain the high frequency of remote mutations observed in evolutionary trajectories (52).

2 6 31

Simon Mathis @simonmathis.bsky.social · Nov 26

Congratulations! This is very exciting news

Reposted by Simon Mathis

Patrick Kidger @patrickkidger.bsky.social · Nov 26

✨ wheeeeee we raised a series B

AI-powered automated protein optimization: matches or exceeds human performance, works without human intervention. It's pretty cool :D

(PS if you're seriously good at ML eng then we're hiring)

www.cradle.bio/blog/series-b

Cradle – Cradle raises $73M Series B to Put AI-Powered Protein Engineering in Every Lab

News, updates, tutorials, and more from the makers of Cradle

www.cradle.bio

2 3 82

Simon Mathis @simonmathis.bsky.social · Nov 26

Amazing! Congratulations to you and the cradle team 🙌

Simon Mathis @simonmathis.bsky.social · Nov 26

Haha where can I order an edition of this? :D

Simon Mathis @simonmathis.bsky.social · Nov 25

As added bonus this allows you to use github copilot or cursor's copilot directly in pymol

Simon Mathis @simonmathis.bsky.social · Nov 25

Check it out here:
github.com/Croydon-Brix...

GitHub - Croydon-Brixton/pymol-remote: Send data to and from pymol from a remote server (e.g. a cluster running deep learning workflows)

Send data to and from pymol from a remote server (e.g. a cluster running deep learning workflows) - Croydon-Brixton/pymol-remote

github.com

Simon Mathis @simonmathis.bsky.social · Nov 25

My workflow:
I use GPUs on my university's cluster to run models, etc.
When I want to look at my designs, I open a pymol session on my laptop, log into the uni vpn and send the structures to my local remote from an interactive session on the cluster.

1 1