Google DeepMind’s New AI Model Predicts How Every Life Molecule Will Behave

Why Trust Techopedia
Key Takeaways

Google DeepMind and Isomorphic Labs have unveiled AlphaFold 3, an AI model that predicts both the structures and interactions of every life molecule.

The follow-up to 2020’s AlphaFold 2, which focused on predicting protein structures, can also model DNA, RNA, ligands, and other organic elements.

AlphaFold 3 is built on a “next-generation” framework with a new version of the Evoformer deep learning module from version 2. Once the AI receives input molecules, it uses a diffusion network to generate a 3D structure from an initial cloud of atoms. It can also model chemical changes to these molecules that might cause disease.

The new approach is “at least” 50% more accurate than earlier prediction techniques for interactions between proteins and other molecules, Google DeepMind said. Accordingly, the company shared multiple examples where the AI model was a “near-perfect match” for the real molecule, including a cold virus and an enzyme.

Google DeepMind has promised to make AlphaFold 3 accessible and responsible. A free AlphaFold Server beta tool lets scientists model structures for non-commercial research using most of the tool’s features. The company added that it had conducted “extensive assessments” to both limit risks and share the advantages with society at large. Education efforts and research partnerships would also help speed up adoption, the firm said.

The implications of AlphaFold 3 are potentially significant. The AI molecule prediction could illustrate many life processes, according to Google DeepMind. That, in turn, could help develop better virus treatments, more biorenewables, and disease-resistant crops. It should also further genomics research as a whole.

The full effect of AlphaFold 3 might take years to arrive. There are also concerns powerful AI models might fall into the wrong hands or otherwise require restrictions that limit their potential. However, Google DeepMind was keen to note that AlphaFold 2 has already been used to advance malaria vaccines and other real-world breakthroughs. In theory, the third iteration might only expand on those discoveries.