DeepMind, the artificial intelligence research lab owned by Google, has developed an improved version of its AI model called AlphaFold 3. This new version not only predicts the structure of proteins, as previous versions did, but it can also predict the structure of “all life’s molecules,” including DNA, RNA, and smaller molecules called ligands. This expanded capability will benefit researchers in fields such as medicine, agriculture, materials science, and drug development by allowing them to test potential discoveries. AlphaFold 3 has shown a 50 percent improvement in prediction accuracy compared to its predecessor, AlphaFold 2.
AlphaFold 3 works by using a diffusion method to generate 3D models of molecular structures. Researchers input a list of molecules they want to combine, and the model uses diffusion to create a new structure. This method is similar to the one used by AI image generators like Stable Diffusion. Isomorphic Labs, a drug discovery company founded by DeepMind CEO Demis Hassabis, has been using AlphaFold 3 for internal projects and has already seen improvements in their understanding of new disease targets. DeepMind is also making the research platform AlphaFold Server available to some researchers for free, allowing them to generate biomolecular structure predictions regardless of their access to computing power.
In order to ensure the responsible deployment of AlphaFold 3, Google is working with the scientific community and policy leaders. They have addressed concerns from biosecurity experts who believe that AI models like AlphaFold 3 could potentially lower barriers for threat actors to design and engineer harmful pathogens and toxins. DeepMind has worked with domain experts, biosecurity specialists, and industry partners to identify and mitigate risks associated with the use of AlphaFold 3. The company is committed to ensuring the safe and ethical use of their technology.
DeepMind CEO Demis Hassabis believes that AlphaFold 3 represents a significant advancement in using AI to understand and model biology. The model’s ability to predict the structures of various molecules beyond proteins opens up a wide range of possibilities for scientific research and discovery. Isomorphic Labs is one example of a company that has already benefited from using AlphaFold 3 in their drug discovery projects. By making AlphaFold Server available to researchers for free, DeepMind is further enabling the scientific community to leverage the power of AI in their work.
Overall, the introduction of AlphaFold 3 by DeepMind represents a significant leap forward in the field of structural biology. The model’s improved accuracy and expanded capabilities make it a valuable tool for researchers in various fields who are looking to better understand and manipulate molecular structures. By working with the scientific community to address potential risks and ensure responsible use of the technology, Google is demonstrating its commitment to ethical AI development. AlphaFold 3 has the potential to revolutionize scientific research and discovery in areas such as medicine, agriculture, materials science, and drug development.