Impact of SMILES Notational Inconsistencies on Chemical Language Model Performance １．This study systematically investigates how inconsistencies in SMILES notation—specifically grammatical and stereochemical variations—affect chemical language models (CLMs), with important

Happening today at **2PM** ET

Multiobjective learning and design of bacteriophage specificity １．This study presents a landmark demonstration of multiobjective machine learning to design bacteriophages with tunable infectivity, precise host specificity, and broad generality, optimizing over 26 complex tasks

🎉Congrats to Nate for his preprint, jointly with Vatsan Raman, where training on multifunctional data enabled precise control over designing phages with complex infectivity & specificity. Big shifts in host targeting come from just a few mutations! tinyurl.com/yc4wtn8h

It’s finally happening!!! Diffusion is so much more satisfying than autoregressive for protein & DNA sequences that don’t really have directionality 🥹 Just we wait for this to empirically land & replace BERT (ie one-step diffusion) for protein foundation models 👀

> duke gets new H200s > Kapil Devkota signs up to be a beta tester > scales up raygun data 20x > raygun zero-shot is now much better > reconstructs arbitrary seqs to >95% identity > so you typically won't need to fine-tune > we also improved the api Enjoy!

Sam Ghali, M.D. Acetaminophen (aka Tylenol) can cause liver failure and death if unintentionally overdosed which can happen fairly easily if not aware. Do not take more than 4 grams in 24 hours which is actually only 8 extra strength pills !

Polly Fordyce at #VariantEffect25

See Isabelle Zane’s amazing artwork summarizing each speaker at the #VariantEffect25 - crazy how well each drawing captures the main messages so beautifully, and for such diverse talks😍! Kresten Lindorff-Larsen

New online! Microbial interactions and ecology in fermented food ecosystems nature.com/articles/s4157…

Deep learning–guided design of dynamic proteins Science Magazine １．This work introduces a general deep learning–guided method for de novo design of dynamic proteins that switch between defined conformational states, mimicking mechanisms seen in natural signaling proteins such

The W-S lab is seeking to hire a research scientist! This person will lead wet-lab innovations in our team as we combine AI and experiments to understand and design dynamics! Thanks for sharing/RTs! jobs.wisc.edu/jobs/scientist…

ChatGPT now can analyze, manipulate, and visualize molecules and chemical information via the RDKit library. Useful for scientific work across health, biology, and chemistry.

Calling RDKit directly from ChatGPT seemed really cool, except my first attempt gave terrible hallucination. The code that was executed invented a SMILES string to load into RDKit. When I draw that hallucinated SMILES, it looks nothing like my requested Alvespimycin, a macrocyle.

If you're curious how the first 5 amino acids of newly generated protein N-termini can affect protein stability in E. coli, then check out our latest preprint We find new patterns of exceptions to the usual N-end rule Kudos to Sunny Sen & co-authors biorxiv.org/content/10.110…

RareFold predicts and designs proteins with noncanonical AAs. With EvoBindRare, we designed linear & cyclic peptide binders with high affinity & novel binding modes, validated in wet lab. A step toward next-gen therapeutics. 📄 biorxiv.org/content/10.110… 💻 github.com/patrickbryant1…

There is this particular strain of argumentation called the Gish gallop where a bunch of poor quality and unconnected points are used to argue something, and it’s really hard to argue against. Trump admin uses this a lot. But let’s start at the high level. American universities

This 🧵 has a much better, more eloquent response to VP Vance than the rant-y style tweet I was about to send out. Read this⬇️

Memorial Day 2025 Freedom isn’t free.

This… this has to be a parody account right…? Not the real secretary of HHS?