Advanced Computing in the Age of AI | Sunday, May 26, 2024

Google DeepMind’s New AlphaFold Model Poised to Revolutionize Drug Discovery 

Google DeepMind, an Alphabet subsidiary focusing on AI, ML, and neuroscience research, has unveiled an improved version of its AI model that predicts the structure and interactions between biological molecules with unprecedented accuracy. 

The upgraded AlphaFold 3 will enable researchers to test potential discoveries in medicine, materials science, drug development, and other fields. The details of this groundbreaking innovation, which has the potential to significantly accelerate biological research, were published in the journal Nature. 

AlphaFold was developed in 2021 in collaboration with London-based drug developer Isomorphic Labs to provide drug researchers with a powerful tool to predict protein structures. With the latest AlphaFold model, researchers can predict the structures of almost all biological molecules and model the interactions between those molecules. This includes interactions between the proteins, RNA, DNA, ligands, and other molecules. 

These new capabilities could hold the key to unlocking new drug discoveries. Previous models solved the problem of predicting the three-dimensional structure of a protein from its amino acid structure, but that had limited utility for drug discovery. With the upgraded AlphaFold 3 model, researchers can now predict how a drug binds with protein, offering more potential for drug discovery. 

“It's a big milestone for us today, announcing AlphaFold 3,” said Demis Hassabis, CEO of Google DeepMind, at a briefing on May 7 announcing the breakthrough. “Biology is a dynamic system and you have to understand how properties of biology emerge through the interactions between different molecules in the cell. You can think of AlphaFold 3 as our first big step towards that.”

Researchers used the previous AlphaFold models as a starting point for drug discovery before pursuing other methods. However, with the better accuracy predictions of AlphaFold 3, researchers now have access to a significantly more powerful tool. DeepMind claims that the new model delivers a 50% improvement in prediction accuracy compared to previous models. 

In addition to the AlphaFold AI models, DeepMind has made several other breakthroughs using AI. In 2022, GoogleMind’s new tool helped predict structures for 2.2 million new materials, out of which 700 went on to be created in a lab. 

In 2023, GoogleMind delivered a new model for weather predictions with unprecedented accuracy. The AlphaFold models have been instrumental in groundbreaking research in several other fields including NLP, healthcare, robotics, and mathematics. 

AlphaFold 3 will be made available via the cloud for researchers to access for free. As of now, DeepMind is not releasing the software as open source. This is a major change in approach as AlphaFold 2 was released as open-source to allow researchers to look under the hood to gain a better understanding of how it worked. However, there are no current plans to release the full code. 

DeepMind will only release a public interface for the model called AlphaFold Server. This model can only be used for noncommercial purposes and has certain limitations on which molecules can be experimented with. According to DeepMind, the limited UI will help lower technical barriers to using the model, enabling users of less technical knowledge to also benefit from AlphaFold. A more accessible model may be released in the future. 

There is no denying that the introduction of AlphaFold 3 is a big leap for drug discovery, and it has a lot of potential, however, it’s still the first step in a long journey toward using AI to understand the possible interactions of biological structures and molecules in nature. 

AlphaFold 3 currently relies on publicly available lab test data sets for its training, and will likely need more advanced training data to achieve its full potential. Ideally, the model needs large, diverse, and multimodal data to adapt to a wide range of biological contexts to learn faster and keep improving its accuracy. 

Related Items 

DeepMind’s AI Algorithm Brings Science One Step Closer to Fusion 

Google, DeepMind Target Machine Learning at Wind Energy 

AI Researchers Issue Warning: Treat AI Risks as Global Priority 

 

EnterpriseAI