PocketGen: Generating Full-Atom Ligand-Binding Protein Pockets
- Co-designs the residue sequence and full-atom structure of protein pockets for binding
- Uses a bilevel graph transformer to model multi-granularity (atom and residue/ligand level) and multi-aspect (intra-protein

I've been playing around with AlphaFold3 since the server was released and, like others have noted, I'm seeing a lot of odd pathological behaviors. For instance, when given 2 copies of this particular protein (+lipids), it more or less just superimposes the structures 1/n

Thank you for organizing, Xiaoxin He!

The release of AlphaFold 3 by Google DeepMind and Isomorphic Labs is a big leap in the prediction of a wider universe of biomolecules.

But, its usefulness for drug discovery, and impact on trial failure rates, is a big open question.

statnews.com/2024/05/08/dru… via STAT

Join me on Sat 12-2pm at the MLGenX Workshop @ ICLR 2024 poster session to talk about PDGrapher, our work on predicting therapeutic perturbations:
openreview.net/forum?id=Zcw2O…

I’m at #ICLR2024 Thurs to Sat - reach out to talk (graph) ML + Bio or about our Prescient Design team!

🦾 AI for Science: Scaling in AI for Scientific Discovery, co-organized by HDSI Faculty Affiliate Marinka Zitnik, will feature speakers recognized for their understanding of scaling’s impact on AI for Science: ai4sciencecommunity.github.io/icml24.html

⏳ Abstract Submission Deadline: May 23, 2024

Exciting new partnership with key priorities, including AI, to accelerate progress in biology through models that leverage data about genetics, patient outcomes, and protein shapes and use these models to provide deep insights into drug design

Focus on deployment & delivery:

💊

That's a wrap on this semester's DBMI at Harvard Med BMI702 Intro to #Biomedical #AI class! 🎓

Explore the syllabus and course materials, open to all: zitniklab.hms.harvard.edu/BMI702/ Harvard Medical School Harvard Data Science Initiative Kempner Institute at Harvard University

To get started run: 'pip install spectrae'

To help you get started we provide tutorials for the following datasets:
1. Deep mutational scan datasets
2. Sequence datasets
3. Single cell perturbation datasets
4. Small molecule datasets
...and more coming soon!

The SPECTRA package generates a series of splits with decreasing train-test similarity. Evaluating your models on these splits will give a better understanding of model generalizability. Read the preprint for more info on how this works:

biorxiv.org/content/10.110…

Introducing the SPECTRA python package!

github.com/mims-harvard/S…

This package implements the spectral framework for model evaluation. All you need to get started is (1) a model, (2) a dataset, and (3) a definition of sample to sample similarity!

A lot of LLM benchmarks don't properly test out of distribution behavior. As uses of LLMs for science increase, we need benchmarks that actually check generalization beyond training data. My experience so far that LLMs are pretty weak outside training distribution, but can have