Isomorphic Labs, Google DeepMind introduce new AI molecule predicting model

Earlier this week, researchers at Google DeepMind and Isomorphic Labs introduced AlphaFold 3, a structure and biomolecule predicting model, sharing their findings in a paper published in Nature.

According to the team, AlphaFold 3 demonstrates a 50% improvement in accuracy over existing methods for modeling interactions between proteins and other molecules, with some interactions showing double the previous accuracy.  AlphaFold 3 is available for scientific use through the newly launched AlphaFold Server, which provides free access to most of its features.

The model's potential is already being leveraged by Isomorphic Labs to accelerate the development of drugs. Building upon its predecessor, AlphaFold 2, AlphaFold 3 expands its capabilities by including training cover all molecular types, making it the first AI system to outperform traditional physics-based tools in biomolecular structure prediction.

In January 2024, Alphabet's Isomorphic Labs entered into two major drug discovery collaborations with Eli Lilly and Novartis, potentially totaling nearly $3 billion in value. With Eli Lilly, Isomorphic Labs will use its advanced AlphaFold technology and substantial computing resources to develop small molecule therapeutics for multiple targets. This agreement includes an initial payment of $45 million to Isomorphic, with potential earnings up to $1.7 billion based on achieving specific milestones.

Isomorphic also partnered with Novartis to leverage its AI technology to discover therapeutics aimed at three undisclosed targets. Novartis will provide Isomorphic with an upfront payment of $37.5 million and will also cover certain research expenses. Further, Isomorphic could earn up to $1.2 billion in milestone payments.