Google DeepMind has finally open sourced the AlphaFold 3 model, making its training weights accessible to academic researchers and scientists—for non-commercial use only.
Check out the model here.
“The AlphaFold 3 model code and weights are now available for academic use. We at Google DeepMind are excited to see how the research community continues to use AlphaFold to address open questions in biology and new lines of research,” said Google DeepMind’s Pushmeet Kohli, as promised six months ago with the announcement to expand AlphaFold 3’s accessibility for the scientific community.
Origin of AlphaFold 3
In May 2024, Google DeepMind released AlphaFold 3, a game changing protein folding model that predicts with 50% better accuracy. It is capable of predicting the structure and interactions of all biological molecules, including proteins, DNA, RNA, and ligands.
Founded by nobel prize winner Demis Hassabis in 2021, Isomorphic Labs—a sister company of Google DeepMind—is looking to revolutionise drug discovery with AI, potentially building a multi-$100 billion business by accelerating research and improving clinical trial success. “AI-designed drugs would probably be available in the next couple of years,” he said.
“I hope to achieve both (commercial success and societal benefits) with Isomorphic and build a multi-100 billion dollar business. I think it has that potential,” said Hassabis without delving into the specific timeline.
The brains behind Isomorphic also include tech veteran Miles Congreve, serving as chief scientific officer, who contributed to the design of 20 clinical-stage drugs and co-invented Kisqali (Ribociclib), a marketed breast cancer treatment.
Also noteworthy is Sergei Yakneen‘s contribution, who is the chief technology officer with over two decades of expertise in engineering, machine learning, product development, and life sciences and medicine research.
The company recently announced key partnerships with two of the world’s largest pharmaceutical companies — Eli Lilly & Co. and Novartis AG. The deals are said to have a combined value of close to $3 billion.
Predicts 200 Mn Protein Structures
Two months ago, Google DeepMind, in collaboration with Isomorphic Labs, predicted over 200 million protein structures using AlphaFold. It achieved this by training the model with nearly 100,000 known proteins, driving significant breakthroughs in drug discovery by targeting previously intractable biomedical challenges.
The model can predict the 3D structure of proteins with incredible accuracy, aiming to design drugs that target specific proteins, unlocking treatments for diseases that were previously untreatable.
Recently, Google DeepMind also launched AlphaProteo, an AI system that generates novel proteins designed to bind to specific target molecules poised to significantly advance research in drug design, disease understanding and other health applications.
DeepMind is not the only active player in the market, ESMFold, Meta’s protein-folding model, has also predicted about 772 million protein structures.
Other protein prediction model include RoseTTAFold, OmegaFold, I-TASSER, Phyre2, SWISS-MODEL, Robetta, HHPred, AlphaFold-Multimer, and CollabFold.
