AlphaFold 3 Release: Breaking Doubts, Rewriting New Drug Development!

To share a very interesting experience, just yesterday I swiped a video of Dr. Yan Ning questioning the reliability of AlphaFold 2 in the year 2021, and today I saw the latest breakthrough of AlphaFold 3 published in the journal Nature and immediately received extensive media attention.

So when I read the Nature article, I focused on whether AlphaFold 3 solved Dr. Yan Ning's problem, and the results showed that AlphaFold 3 has indeed made breakthroughs in structural biology, both in terms of being able to jointly predict the structure of complexes including proteins, nucleic acids, small molecules, ions, and modifying residues, and in terms of being more accurate in protein-ligand interactions, protein-nucleic acid interactions, and antibody-antigen prediction. and antibody-antigen prediction.

Since its introduction in 2021, AlphaFold2, an artificial intelligence (AI) protein structure prediction tool, has revolutionized the field of bioscience with astonishing accuracy. Scientists have used AlphaFold2 to resolve organelle structures, explore drug molecular design, and even map the macroscopic nature of known proteins.
However, John Jumper, the leader of AlphaFold2, is still constantly asked if the tool can go further, such as predicting the shape of functionally modified proteins or the structure of their interactions with cellular components such as DNA and RNA.

The latest version of AlphaFold, AlphaFold3, aims to achieve this goal by empowering scientists to predict the structure of proteins when they interact with other molecules, providing a whole new way of thinking about drug development.

The researchers found that AlphaFold3 significantly outperforms existing software tools in predicting the structure of proteins and their binding partners. For example, scientists interested in drug discovery often use "docking" software to simulate the binding of compounds to proteins, which often requires the experimental structure of the protein, and AlphaFold3 already outperforms two commonly used docking software programs and another AI-based tool, RoseTTAFold All-Atom, in terms of accuracy. AI-based tool RoseTTAFold All-Atom.
Frank Uhlmann, a biochemist at the Francis Crick Institute in London, gained early access to AlphaFold3 and raves about its capabilities: "It's nothing short of revolutionary, and it's going to make research in the field of structural biology much more accessible."
Uhlmann's team used AlphaFold3 to predict the structure of DNA-interacting proteins involved in genome replication and verified through mutation experiments that the predictions are often very accurate.
"AlphaFold3's structural prediction ability is impressive," says David Baker, a computational biophysicist at the University of Washington. It performed better than the RoseTTAFold All-Atom his team developed.
However, unlike DeepMind's treatment of AlphaFold2, which will be publicly available without restriction in 2021, AlphaFold3 is currently limited to research for non-commercial use through the DeepMind website.
In particular, it should be noted that despite the new milestone breakthrough, AlphaFold3 still has many flaws.
But just as in the early stages of automobile development, they were not always superior to horse-drawn carriages due to cost, infrastructure, and reliability constraints; over time, with technological advances and increased productivity, automobiles gradually surpassed horse-drawn carriages as the dominant mode of transportation.
Similarly, in the early stages of gunpowder rifles, bows and arrows were usually superior in terms of operational efficiency, continuous combat capability, and accuracy; but over time and with advances in gunpowder technology, gunpowder rifles gradually improved in performance, eventually replacing bows and arrows in the vast majority of military endeavors.
The same is true for the new technological breakthroughs represented by AlphaFold3, and all we can do is to follow Dr. Yan Ning's words: true researchers embrace technological advances and make the best use of various technologies to explore and answer the questions they are interested in.

【1】 https://www.nature.com/articles/s41586-024-07487-w

【2】 https://www.nature.com/articles/d41586-024-01383-z