Google open-sources Nobel Prize-winning chemical model Alphafold-3: will shorten new drug, vaccine development at the limit

Google DeepMind made AlphaFold3 open source! AlphaFold3 has already gained 1.7k stars since it was first open sourced.

At this point, scientists can download the AlphaFold3 code and use this AI protein prediction tool for non-commercial purposes. It is worth noting that only scientists with an academic background can access the training weights upon request.

It is reported that the main features of AlphaFold-3 include the ability to make structural predictions of a wide range of biomolecules, covering proteins, nucleic acids (including DNA and RNA), small molecules, ions, and modifying residues of virtually all molecule types in the Protein Data Bank (PDB).

GitHub address: https://github.com/google-deepmind/alphafold3

In May of this year, Google DeepMind and Isomorphic Labs launched AlphaFold 3, a revolutionary artificial intelligence (AI) model that successfully predicts the structures and interactions of all life molecules (proteins, DNA, RNA, ligands, etc.) with unprecedented accuracy, and the related research paper was published in the authoritative scientific journal Nature.

Previously, Google DeepMind did not provide code or model weights for AlphaFold3, which drew criticism from scientists who argued that the move undermined reproducibility.

Now, the open source of AlphaFold3 will further change our understanding of the biological world and drug discovery, and thus usher in a new era of AI cell biology.

"We're excited to see what people do with it," said AlphaFold team leader John Jumper. Last month, he and Google DeepMind CEO Demis Hassabis were jointly awarded the 2024 Nobel Prize in Chemistry for their work on AlphaFold.

As described in the research paper, AlphaFold 3 found that protein interactions with other molecule types improved by at least 50% compared to existing prediction methods, and even doubled the prediction accuracy for some important interaction classes.

To capitalize on the potential of AlphaFold 3 for drug design, Isomorphic Labs has partnered with pharmaceutical companies to apply it to real-world challenges and ultimately develop new therapies for some of the most devastating diseases affecting humanity.

Although AlphaFold 3 was not previously open-sourced, according to DeepMind's head of AI science, Pushmeet Kohli, several replicated results of AlphaFold3 have already appeared in the industry, suggesting that the model can be replicated, even without open-source code.

Indeed it is also true that in the last few months several companies, relying on the specifications (i.e. pseudo-code) described in the original paper, have launched open source protein structure prediction tools based on AlphaFold3.

Mohammed AlQuraishi, a computational biologist at Columbia University in New York, said a major limitation of the models is that, like the previously unopened-source AlphaFold3, none of them are licensed for commercial applications such as drug discovery.

San Francisco-based Ligo Biosciences has released an unrestricted version of AlphaFold3; other teams are working on unrestricted versions of AlphaFold3: AlQuraishi hopes to release OpenFold3, a fully open-source model, by the end of this year, which will allow pharma companies to retrain their own versions of the model using proprietary data (such as protein structures bound to different drugs) to improve performance. protein structures that bind to different drugs) to retrain their own versions of the model, thereby improving performance.

The open-source nature of AlphaFold3's predecessor model, AlphaFold2, has sparked a great deal of innovation from other scientists. For example, the winners of a recent protein design competition used the AI tool to design new proteins capable of binding to cancer targets; and a team used the tool to identify a key protein that helps sperm attach to egg cells.

Jumper can't wait for such surprises to come after sharing AlphaFold3 - even if they don't always turn out to be fruitful. "People are going to use it in weird ways," he predicts, "and sometimes it's going to fail, and sometimes it's going to work."