Alex Morehead (何聪)

Hopper Postdoctoral Fellow @BerkeleyLab. Prev: #MachineLearning & #CompBio PhD @Mizzou; Research Intern @ Profluent & Absci. #DeepLearning & #GenerativeModels.

[Alex Morehead (何聪) @MoreheadAlex]

Introducing Zatom-1, the first end-to-end, fully open-source foundation model for 3D chemistry! This was a great collaborative effort with many brilliant scientists. I'm grateful to have played a small part.

Paper: https://t.co/QXRCwnCFN8
Code: https://t.co/OC4JLhwsiR

[Alex Morehead (何聪) @MoreheadAlex]

@cohenrap @lazar_atan @frimpongboadu1 @joelselvaraj95 @AlexanderTong7 @ask1729 @jianlincheng Apologies, just got access to a (free) shareable link: https://t.co/eFtcnrCTip

[Alex Morehead (何聪) @MoreheadAlex]

A review of flow matching and its role in CompBio/virtual cell research, now in Nature Machine Intelligence!

Congrats, team! (@lazar_atan, Akshata Hegde, Yanli Wang, @frimpongboadu1, @joelselvaraj95, @AlexanderTong7, @ask1729, @jianlincheng)

Paper: https://t.co/sc8WnXhwFA

[Alex Morehead (何聪) @MoreheadAlex]

Excited to announce that PoseBench is now published in Nature Machine Intelligence!

Paper: https://t.co/JvTwNJISeX
Code: https://t.co/0jtJIo74PW

[Alex Morehead (何聪) @MoreheadAlex]

Interested in learning more about how flow matching has begun to advance bioinformatics and computational biology? And how it has already started making strides towards the development of an AI-based virtual cell?

Paper: https://t.co/haI8aSPdvN
Code: https://t.co/BEJAuVw09i

[Minjia Zhang @_Minjia_Zhang_]

🔬Interested in training AlphaFold3 faster, at scale, and beyond NVIDIA GPU? Now you can.

AlphaFold3 is a major leap in biomolecular modeling, but behind the scenes, it introduces severe system bottlenecks:

🧠 2D EvoAttention spikes memory usage

📉 Retrieval-augmented training pipeline causes long GPU idle time

⛔ Frequent but memory-intensive ops slow everything down

Today, I'm excited to announce MegaFold, a fully open-source system to make AlphaFold3 training fast, scalable, and cross-platform on both NVIDIA and AMD GPUs.

MegaFold delivers:

⚡ Up to 1.73x / 1.62x faster training on NVIDIA H100 / AMD MI250

🧬 Up to 1.35× longer sequences compared to PyTorch baseline

Key features:

🚀 Memory-Efficient EvoAttention via portable Triton kernels

💡 Ahead-of-Time Caching to eliminate GPU idle time in retrieval pipelines

🔗 DeepFusion for reducing overhead of small but frequent memory-intensive AF3 ops

📘 Project page: https://t.co/sSaxDtBT1O

📄 Paper: https://t.co/2sCArFoq82

💻 Code: https://t.co/gn7HR5kcQC

🤝 MegaFold is developed in collaboration between UIUC SSAIL Lab and researchers from University of Missouri and Lawrence Berkeley National Laboratory.

Kudos to the brilliant team: Hoa La, Ahan Gupta, Alex Morehead, Jianlin Cheng

#AlphaFold3 #AI #ProteinFolding #Bioinformatics #AMD #Triton #CrossPlatform #OpenSource

[Alex Morehead (何聪) @MoreheadAlex]

Delighted to announce that FlowDock has been accepted to ISMB 2025. See you all in Liverpool!

Paper: https://t.co/zvl7wbHVB8
Code: https://t.co/F6t9uW2w7V

[Alex Morehead (何聪) @MoreheadAlex]

v0.6.0 of PoseBench is now available, featuring (1) results for AlphaFold 3, the new PLIF-WM metric, and (3) the new DockGen-E dataset of challenging docking targets. See the GitHub release below for more details.

Paper: https://t.co/rKCnwvD8lC
Code: https://t.co/0jtJIo74PW

[Alex Morehead (何聪) @MoreheadAlex]

Excited to release FlowDock, an all-atom flow matching model for generative protein-ligand docking and affinity prediction (ranked as a top method in CASP16)!

Paper: https://t.co/zvl7wbItqG
Code: https://t.co/F6t9uW33Xt

[Alex Morehead (何聪) @MoreheadAlex]

PoseBench v0.5.0 is now released, featuring (1) docking results with AlphaFold 3's predicted protein structures, (2) Chai-1's benchmarking results, and (3) support for running exhaustive HPC benchmarking sweeps. 🧪

Paper: https://t.co/rKCnwvD8lC
Code: https://t.co/0jtJIo74PW

[Alex Morehead (何聪) @MoreheadAlex]

At @icmlconf this week where I'm presenting PoseBench at the AI4Science workshop as a spotlight. I'll also give an oral presentation on RNA-FrameFlow at the SPIGM workshop (AI4Science spotlight as well!) on behalf of many amazing collaborators including @rishabh16_ and @chaitjo.

[Alex Morehead (何聪) @MoreheadAlex]

Introducing PoseBench, the first deep learning (DL) benchmark for practical protein-ligand docking, which provides actionable insights for the development of future docking methods. 🧵
Paper: https://t.co/rKCnwvD8lC
Code: https://t.co/0jtJIo74PW

[Rishabh Anand (in SF) @rishabh16_]

Unfortunately can't join in-person @icmlconf 🇦🇹 but our awesome co-author @MoreheadAlex will be there!!!

Check out our Oral presentation @ SPIGM Workshop on 26 July and Spotlight poster @AI_for_Science Workshop on 27 July ✨🥳

See our poster + schedule below 👀👇🏻

[Alex Morehead (何聪) @MoreheadAlex]

Examples of pocket-specific 3D molecules generated by GCDM:

[Alex Morehead (何聪) @MoreheadAlex]

I'm excited to announce that GCDM for 3D molecule diffusion generation and optimization is now published in Nature @CommsChem!

Paper: https://t.co/pnqRVmnjTe
Code: https://t.co/uu0CiuaKsK

[Alex Morehead (何聪) @MoreheadAlex]

GCDM also generates significantly more PoseBusters-valid 3D molecules in target protein pockets.

[Alex Morehead (何聪) @MoreheadAlex]

GCDM enables out-of-the-box (reliable) property and stability-specific optimization of existing 3D molecules.

[Alex Morehead (何聪) @MoreheadAlex]

GCDM generates more than twice as many PoseBusters-valid large (i.e., GEOM-Drugs-sized) 3D molecules compared to existing methods.

[Alex Morehead (何聪) @MoreheadAlex]

GCDM generates 3D molecules with specific molecular properties more accurately and stably compared to existing methods.

[Alex Morehead (何聪) @MoreheadAlex]

Pocket-only docking results for the PoseBusters Benchmark dataset:

[Alex Morehead (何聪) @MoreheadAlex]

Introducing PoseBench, the first deep learning (DL) benchmark for practical protein-ligand docking, which provides actionable insights for the development of future docking methods. 🧵
Paper: https://t.co/rKCnwvD8lC
Code: https://t.co/0jtJIo74PW

[Alex Morehead (何聪) @MoreheadAlex]

New benchmark results for the DockGen dataset from the recent DiffDock-L paper: