Zhonglin Cao

[Kieran Didi @DidiKieran]

📢 We’re launching Proteina-Complexa — and after the Jensen keynote mention, we definitely had to post this thread now ;)
Atomistic binder design with generative pretraining + test-time compute, plus large-scale wet-lab validation.
Project page: https://t.co/aT8Lz2VhSJ
🧵 1/n

[Karsten Kreis @karsten_kreis]

📢📢 Proteina-Complexa 📢📢

Atomistic Binder Design with Generative Pretraining and Test-Time Compute + Experimental Validation at Scale

⭐️ Project page (https://t.co/8GONsGrard) for:

📜 Method paper (ICLR 2026 Oral)
🧬 Wet lab paper
🛠️ Code & models
📁 Data

🧵 Thread

(1/n)

[Kieran Didi @DidiKieran]

Too many REPA / RAE / representation alignment papers lately?
I was lost too, so I wrote a blog post that organizes the space into phases and zooms in on what actually matters for general/molecular ML.
Curious what folks think - link below!

🔗 Blog: https://t.co/6aJf8DCWTa

[Arash Vahdat @ArashVahdat]

🚀 NVIDIA unveiled 4 open model families: Nemotron for digital AI, Cosmos for physical AI, Isaac GR00T for robotics & Clara for biomedical AI.

🔥Thrilled that GenMol, ReaSyn & La-Proteina under Clara were developed by my team & BioNemo R&D jointly.

https://t.co/fmW3o75CV9

[Zhonglin Cao @ZhonglinJC]

Why do rotational data augmentation when you can train model to learn the flow to the average of all rotations? Check out AvgFlow, the work originated from the brilliant idea of @mario1geiger!

[NVIDIA Healthcare @NVIDIAHealth]

What is AvgFlow?

It is a model training and inference framework for accelerating flow-based models that generate 3D molecular conformers.

Uses SO(3)-Averaged Flow to accelerate model training (fewer epochs needed and eliminating the need of rotation augmentation) and reflow + distillation to accelerate model inference (fewer inference steps), yielding faster convergence and better performance across architectures while maintaining accuracy.

🔗 ⬇️

[NVIDIA Healthcare @NVIDIAHealth]

What is AvgFlow?

It is a model training and inference framework for accelerating flow-based models that generate 3D molecular conformers.

Uses SO(3)-Averaged Flow to accelerate model training (fewer epochs needed and eliminating the need of rotation augmentation) and reflow + distillation to accelerate model inference (fewer inference steps), yielding faster convergence and better performance across architectures while maintaining accuracy.

🔗 ⬇️

[NVIDIA Healthcare @NVIDIAHealth]

🧬La-Proteina🧬

The first generative model demonstrating accurate co-design of fully atomistic protein structures (sequence + side-chains + backbone) at scale, up to 800 residues, with state-of-the-art atomistic motif scaffolding performance - has just made its code open-source!

Learn more 🧵

[Zhonglin Cao @ZhonglinJC]

Check out La-Proteina!

[Tomas Geffner @tomasgeffner]

Presenting La-Proteina! A new model for scalable, all-atom protein design 🧬 Backbone + sequence + side-chains, indexed and unindexed atomistic motif scaffolding, scalable up to 800 residues, and more…

A thread 🧵

[Zhonglin Cao @ZhonglinJC]

Checkout our work Proteina with beautiful visualization on https://t.co/Pf03SieQ29!

[Karsten Kreis @karsten_kreis]

📢📢 "Proteina: Scaling Flow-based Protein Structure Generative Models"

#ICLR2025 (Oral Presentation)

🔥 Project page: https://t.co/OZNXeFdWRc
📜 Paper: https://t.co/AAk3AySUrz
🛠️ Code and weights: https://t.co/jx8jTRLyag

🧵Details in thread...

(1/n)

[Zhonglin Cao @ZhonglinJC]

@YuyangW95 🤯

[Machine Learning in Chemistry @ML_Chem]

Large-Scale Pretraining Improves Sample Efficiency of Active Learning-Based Virtual Screening #machinelearning #compchem https://t.co/BysK9BwNKa