Ben Fry

Very exciting! I tried out the inference code but the script is current crashing after conformer generation due to missing the AimNet model weights. Did I miss a download link somewhere?

GitHub - isayevlab/LoQI: LoQI: Low Energy QM Informed Conformer Generation LoQI: Low Energy QM Informed Conformer Generation. Contribute to isayevlab/LoQI development by creating an account on GitHub.

The paper represents a paradigm shift by combining machine-learned interatomic potentials (MLIPs) with generative modeling to bypass traditional conformer generation, achieving both higher accuracy and greater efficiency than existing methods. Free & open source: github.com/isayevlab/LoQI

Interested in doing a postdoc at DFCI/Harvard on computationally designing and experimentally characterizing mini-protein binders for biomedical applications? Eric Fischer and I are looking for someone to work in our groups starting asap! Email me or my admin with a CV to apply!

🚨New paper 🚨

Can protein language models help us fight viral outbreaks? Not yet. Here’s why 🧵👇
1/12

Hardware/wetware codesigned data loop VISTA makes use of generative model sampling and synthesis "on chip" on-board by leveraging oligosynthesis setup shown here.

The biggest challenge for AI in biology isn't just models, it's the data used to train them. Standard biological data isn't built for AI. To unlock generative AI for drug discovery, we must rethink how we generate and capture data. 1/

Triple Honours for Franziska Sendker Dr Franziska Sendker, a former doctoral candidate at the Max Planck Institute for Terrestrial Microbiology, has been awarded the Otto Hahn Medal by the Max Planck Society. The medal honours the outstanding achievements of young scientists and comes with a prize of 7,500 euros. In March 2025, she received the Bayer Pharmaceuticals Doctoral Award from the Society for Biochemistry and Molecular Biology (GBM e.V.), presented at the Mosbach Colloquium. Franziska Sendker's research showed that complex protein forms can arise not only through natural selection, but also through random genetic changes.

📢 Congratulations 𝗗𝗿. 𝗙𝗿𝗮𝗻𝘇𝗶𝘀𝗸𝗮 𝗦𝗲𝗻𝗱𝗸𝗲𝗿 for receiving the Otto Hahn Medal and Otto Hahn Award from @maxplanck.de ! 🎉 The honors recognize her exceptional work @georghochberg.bsky.social. 🌟 Exciting times ahead! #MaxPlanck #ResearchExcellence www.mpi-marburg.mpg.de/1511259/2025...

Interesting though that Boltz-2 ipTM seems to rank the affinities /almost perfectly for our point mutants and SAR. Curious to see what others are using to rank designed interfaces 🤔 2/2

Trying out the Boltz-2 affinity prediction on the Exatecan binders we generated with LASErMPNN and NISE. Affinity prediction still clearly has room to improve, but the model seems to be able to identify the highest affinity mutant in this small dataset. Thanks to @gcorso.bsky.social and team! 1/2

from most recent Harvard lawsuit. sums it up pretty succinctly I think

Congrats Liana 🎉

YouTube video by Boston Protein Design and Modeling Club Design of small molecule binding proteins using deep learning

in case you missed the superb seminar that Ben Fry gave back in January, you can now check out the recording youtu.be/IgFgAYQrke4
and the preprint www.biorxiv.org/content/10.1...

We're super excited by the method. We think it can help to rapidly produce binders to small molecules for sensors, antidotes, delivery vehicles, even enzymes. Let us know what you think and please try it out! Finally, shout out to Ben and Kaia for making this all happen!!! 🤩

absorbance data showing that epic protect exatecan from hydrolysis

Lastly, Kaia checked to see if EPIC and its higher affinity mutant are able to protect exatecan from hydrolysis, which is not something serum albumin can do. For a drug that normally hydrolyzes in a few hours, EPIC was able to stabilize the lactone form for days! ✅

co-structure predictors agree with crystal structure but differ at ligand

Since EPIC and exatecan aren't in the PDB, we wanted to see how co-structure predictors do on it. They each get the backbone right but differ at the ligand. The pose is correct but the modeling of the conformer is wonky. AF3 does the best. AF3 is also able to rank affinities via pLDDT of ligand! 😱

crystal structure of EPIC agrees with design

Kaia was able to crystalize EPIC and determine its structure to 2.0 Å resolution. It agreed pretty well with the LASEr design! RFAA had a hard time modeling the lactone ring of the drug, so there is some disagreement there. The lactone is buried as intended, and the goal was to hide it from water 👍

example of neural proofreading to improve affinity

Ben didn't stop there. He wanted to improve affinity of EPIC for exatecan using computation alone. He used LASErMPNN to "proofread" EPIC's sequence using a predicted co-structure as input. LASEr suggested two mutations. Kaia verified that each improved binding 10x. 100x when combined (1 nM Kd)!

binding curves

Kaia Slaw (no bluesky) experimentally tested 4 designs from NISE and 16 from COMBS. All 4 NISE designs bound! The highest affinity binder- which Ben and Kaia call "EPIC" - was pretty tight (0.1 uM Kd). Compared to COMBS (3 of 16 bound, tightest was 10 uM), NISE and LASErMPNN did a much better job!

image of exatecan drug and design pipelines using COMBS and NISE

Ben used NISE and LASErMPNN to design binders to exatecan, an anticancer drug prone to inactivation by hydrolysis. We also used a more "traditional" approach using COMBS and Rosetta to design binders. We could compare the methods head to head.

two proteins that have good predicted structures but only one has a self consistent ligand position

With the new co-structure predictors like RFAA, Boltz-1, and AF3, we can now extend self-consistency into the ligand dimension. And Ben's NISE algorithm maximizes this. Code repo here: github.com/polizzilab/N...

protein sequence design and structure prediction showing some designs agree with the intended structure

We all know in protein design about the goal of self consistency. That is, we want the predicted structure to look like the structure for which we designed the sequence.

the NISE design algorithm iteratively optimizes a starting model through many rounds of sequence design and structure prediction

Ben used LASErMPNN in combination with a protein-ligand co-structure predictor, RFAA, in an iterative algorithm called NISE that refines designs. NISE optimizes the sequence, structure, and ligand conformer together to improve the confidence of both models. It's a neural-network-only algorithm

scatter plot of sequence recovery on a test set showing lasermpnn improves slightly over ligandmpnn

Ben Fry (@benf549.bsky.social) was excited when proteinMPNN came out, which motivated him to train a new gNN called LASErMPNN to design sequences given protein-ligand co-structure. LASErMPNN does pretty well at this! The repo is available and even has the training code! github.com/polizzilab/L...

Zero-shot design of drug-binding proteins via neural selection-expansion Computational design of molecular recognition remains challenging despite advances in deep learning. The design of proteins that bind to small molecules has been particularly difficult because it requ...

Super excited to share a new preprint from our lab on design of small-molecule binding proteins using neural networks! The paper has a bit of everything. A new graph neural network, new design algorithms, and experimental validation. www.biorxiv.org/content/10.1...
🧵🧪

Covalent Modifications to Protein Sequence Retain OXT Atoms · Issue #94 · jwohlwend/boltz Hey Boltz Team, Thanks for taking on this project and for writing such clean code. I was playing around with covalently modified residues and I ran into a bug where the OXT leaving atom is not remo...

I’ve noticed this as well and made a pull request to fix it that hasn’t been merged yet. You can clone the branch I mention here for a fix github.com/jwohlwend/bo...

Awesome to see a fully open source model. Looking forward to using this Gabriele!