Nate

Discovery of a New and Selective HPK1 PROTAC for Enhancing Tumor Immunotherapy through Eliminating GLK Degradation HPK1 is an attractive therapeutic target for tumor immunotherapy. Nevertheless, the formidable challenge selectivity over GLK and limited antitumor efficacy of HPK1 inhibitors and PROTACs impeded thei...

Discovery of a New and Selective HPK1 PROTAC for Enhancing Tumor Immunotherapy through Eliminating GLK Degradation

F.D.A. to Use A.I. in Drug Approvals to ‘Radically Increase Efficiency’ With a Trump-driven reduction of nearly 2,000 employees, agency officials view artificial intelligence as a way to speed drugs to the market.

🧪In a JAMA viewpoint paper, the FDA expresses its intent to use AI for first-pass reviews of drug applications to significantly increase efficiency.

This is premature and we haven't seen the evidence AI is ready for this use case.

Link to the paper: jamanetwork.com/journals/jam...

#MedSky #MLSky

Someone suggested Vancouver yesterday. They were joking, but I'll bring it up next time I see Daphne

On the design space between molecular mechanics and machine learning force fields A force field as accurate as quantum mechanics (QMs) and as fast as molecular mechanics (MMs), with which one can simulate a biomolecular system efficiently eno

Force fields with QM accuracy and MM speed are the theoretical biophysicist's Philosopher's Stone. I used ANIX2 with OpenFF and OpenMM. The results were insightful but at the expense of huge computational costs. #MLFF #biophysics #moleculardynamics

Sharing slides for All-atom Diffusion Transformers

- briefly summarises the big ideas and key takeaways

Link - www.chaitjo.com/publication/...

Man, I gotta say, @aspuru.bsky.social definitely made the right choice to bail on the US way back in 2018. I mean I understood back then, but... prescient decision!!

Incredibly grateful to have helped build the ASAP Discovery Consortium (@asapdiscovery.bsky.social) that enabled @griffen-ed.bsky.social and all the other amazing members of this team to discover a new broad-spectrum coronavirus antiviral!

Chemistry is changed by AI at a super rapid pace! Check our data miner for chemical reactions, MERMaid!

@accelerationc.bsky.social @thematterlab.bsky.social @shixuanleong.bsky.social #chemsky #compchemski #ai #vlm #llm #aiforscience Please follow @thematterlab.bsky.social

What an awesome group effort reporting on the structural dynamics of ~60% of available structures !!

www.nature.com/articles/s41...

not just NIH websites

The next step in trying to dismantle Infectious Diseases research - dismantle @niaidnews.bsky.social - should surprise no one that this a R-led proposal.

An unmitigated disaster for diagnosis, treatment and prevention of pathogens that affect everyone in the US

www.congress.gov/bill/119th-c...

Our first attempts at mechanistic interpretability of Transformers from the perspective of network science and graph theory!

A wonderful collaboration with superstar MPhil students Batu El, Deepro Choudhury, as well as Pietro Liò as part of the Geometric Deep Learning class at @cst.cam.ac.uk

Variational Flow Matching goes Riemannian! 🔮

In this preliminary work, we derive a variational objective for probability flows 🌀 on manifolds with closed-form geodesics, and discuss some interesting results.

Dream team: Floor, Alison & Erik (their @ below) 💥

📜 arxiv.org/abs/2502.12981
🧵1/5

De novo design of transmembrane fluorescence-activating proteins - Nature A study describes the design of de novo ligand-binding transmembrane proteins, demonstrating their specific binding and activation of fluorogenic ligands.

WOW!!
De novo design of transmembrane fluorescence-activating proteins
www.nature.com/articles/s41...

Figure 1 from arXiv preprint https://doi.org/10.1101/2025.01.06.631610 Fig. 1 Espaloma is an end-to-end differentiable molecular mechanics parameter assignment scheme for arbitrary organic molecules. Espaloma (extensible surrogate potential optimized by message-passing) is a modular approach for directly computing molecular mechanics force field parameters FFF from a chemical graph G such as a small molecule or biopolymer via a process that is fully differentiable in the model parameters FNN. In Stage 1, a graph neural network is used to generate continuous latent atom embeddings describing local chemical environments from the chemical graph. In Stage 2, these atom embeddings are transformed into feature vectors that preserve appropriate symmetries for atom, bond, angle, and proper/improper torsion inference via Janossy pooling.54 In Stage 3, molecular mechanics parameters are directly predicted from these feature vectors using feed-forward neural networks. This parameter assignment process is performed once per molecular species, allowing the potential energy to be rapidly computed using standard molecular mechanics or molecular dynamics frameworks thereafter. The collection of parameters FNN describing the espaloma model can be considered as the equivalent complete specification of a traditional molecular mechanics force field such as GAFF38,39/AM1-BCC55,56 in that it encodes the equivalent of traditional typing rules, parameter assignment tables, and even partial charge models. Reproduced from ref. 49 with permission from the Royal Society of Chemistry.

Everything is chaos, but I wanted to share some awesome recent science from the lab that hints at where the future of biomolecular simulation is headed:

Foundation simulation models that can be fine-tuned to experimental free energy data to produce systematically more accurate predictions.

The BioEmu-1 model and inference code are now public under MIT license!!!

Please go ahead, play with it and let us know if there are issues.

github.com/microsoft/bi...

The Continuing Crisis, Part IV

Here’s the latest in the Continuing Crisis series - there will, unfortunately, be more:

As our debut Bluesky post, we’re excited to share our new paper (first author Zachary McCaw) in HGG Advances on scrutinizing the practice of using a ratio trait (numerator / denominator) for GWAS. www.cell.com/hgg-advances...

Very cool work!!

#BPS2025 is so close!
Make sure you are well prepared.

Downlod the BPS Mobile App and use the Desktop Planner to develop an itinerary that misses none of the exciting opportunities of an in-person meeting.

@biophysicalsoc.bsky.social #science #ionchannels #iGluRs

mailchi.mp/biophysics/p...

I’m putting together a post on what’s going on in the trenches in various parts of the NIH. Anyone who would like to share experiences (anonymously!) can reach me at Dblowe.18 on Signal

The variability across the 20 conformations in a typical NMR ensemble in the PDB represents the uncertainty from the lack of experimental data and not the dynamics of the protein in solution

Thanks for coming to my TED talk

Benchmarking gene embeddings from sequence, expression, network, and text models for functional prediction tasks Gene embeddings have emerged as transformative tools in computational biology, enabling the efficient translation of complex biological datasets into compact vector representations. This study present...

"One of the most notable findings from our benchmarks is the superior performance of text-based embeddings across all tasks" although the authors warn this could be due to data leakage