Google DeepMind Unveils Next Generation of Drug Discovery AI Model
AsianFin--Google DeepMind has released a new version of AlphaFold, a landmark tool for predicting protein structures, that puts the artificial intelligence ( AI ) software on a path to make breakthroughs in biology research and bolster a business that Google ’ s AI chief says could be worth north of $100 billion.
AlphaFold, an AI system developed by Google DeepMind that predicts a protein ’ s 3D structure from its amino acid sequence, was initially launched in 2018. In late 2020, the company made a significant advance in molecular biology by using AlphaFold 2 to successfully predict the behavior of microscopic proteins.
AlphaFold3, the newest version of AlphaFold, developed by researchers at DeepMind and its sister company Isomorphic Labs under the leadership of cofounder Demis Hassabis, has successfully mapped the behavior of all molecules essential for life, including human DNA.
The interactions of proteins - from enzymes crucial to the human metabolism, to the antibodies that fight infectious diseases - with other molecules is key to drug discovery and development.
DeepMind said the findings, published in research journal Nature on Wednesday, would reduce the time and resources needed to develop potentially life-changing treatments.
"With these new capabilities, we can design a molecule that will bind to a specific place on a protein, and we can predict how strongly it will bind," Hassabis said in a press briefing on Tuesday. "It's a critical step if you want to design drugs and compounds that will help with disease."
The company also announced the release of the "AlphaFold server", a free online tool that scientists can use to test their hypotheses before running real-world tests.
Since 2021, AlphaFold's predictions have been openly available to non-commercial researchers as part of a database comprising over 200 million protein structures. It has been referenced thousands of times in the work of others.
DeepMind said the new server required less computing knowledge, allowing researchers to run tests with just a few clicks of a button.
John Jumper, a senior research scientist at DeepMind, said: "It ’ s going to be really important how much easier the AlphaFold server makes it for biologists – who are experts in biology, not computer science – to test larger, more complex cases."
Dr. Nicole Wheeler, an expert in microbiology at the University of Birmingham, said AlphaFold 3 could significantly speed up the drug discovery pipeline, as "physically producing and testing biological designs is a big bottleneck in biotechnology at the moment."
AlphaFold 2 accomplished a significant scientific feat by accurately predicting the structure of most proteins solely from their DNA sequence. Subsequently, the company published the system's predicted structures for all 200 million proteins with known DNA sequences, providing free access to scientists through a vast database. Prior to this, only approximately 100,000 proteins had known structural information.
Understanding the shape and configuration of a protein is crucial in comprehending its functionality. However, proteins do not operate independently. Initially, AlphaFold 2 was not designed to forecast how proteins would interact with each other, although scientists quickly devised methods to adapt AlphaFold 2 for making some of these predictions. Additionally, AlphaFold 2 couldn't anticipate protein interactions with other types of molecules, such as DNA, RNA, ligands, and ions, which are present within living organisms. Furthermore, it couldn't predict the interactions among these other molecules. AlphaFold 3 addresses these limitations.
While the system isn't infallible, it signifies a significant advancement in performance. According to assessments conducted by Google DeepMind and Isomorphic, AlphaFold 3 can accurately predict 76% of protein interactions with small molecules, surpassing the previous best predictive software, which achieved 52%. It can predict 65% of DNA interactions, compared to the next leading system's 28%. Moreover, in protein-to-protein interactions, it can predict accurately 62%, more than double what AlphaFold 2 achieved.
Similar to AlphaFold 2, AlphaFold 3 provides a confidence score alongside its predictions, offering scientists insight into the reliability of the system's output. This diminishes the likelihood of the AI model encountering "hallucinations" — believable yet inaccurate outputs — that have affected recent generative AI models.
