Google DeepMind Open Sources AlphaFold 3 AI Model to Help Researchers in Drug Discovery

Google DeepMind has silently open-sourced its frontier artificial intelligence (AI) model that can predict the interaction between proteins and other molecules. Dubbed AlphaFold 3, the large language model is the successor of AlphaFold 2, whose research led to the creators of the large language model (LLM) Demis Hassabis and John Jumper getting the Nobel Prize in Chemistry in 2024. AlphaFold 3 takes the capabilities further with its ability to model proteins' interaction with DNA, RNA, and other smaller molecules which can potentially lead to drug discovery.

Google DeepMind Open-Sources AlphaFold 3 AI Model

Research on protein structures has been one of the major areas of focus in Chemistry. Since the 3D shape and atomic details of proteins are the targets for drugs, discovering new protein structures can often open previously unexplored targets and mechanisms for medical intervention. Put simply, the better we understand protein structures, the more effective medicines can be against various disorders, diseases, and autoimmune disorders.

While Google DeepMind made no announcement about releasing the AlphaFold 3 AI model, it has made the source code and model weights available on GitHub. However, this is only available for academic and research purposes. The source code is available freely under a Creative Commons licence, however, the weights can only be accessed after obtaining direct permission from Google for academic use.

It is believed that if the AI model can correctly highlight how proteins interact with DNA, RNA, and other smaller molecules, researchers will be able to accelerate the manufacturing of new synthetic drugs.

Researchers will also be able to automate work that could have taken them years without any proof of success. AlphaFold 3 comes three years after the release of AlphaFold 2 in 2021. In a study, the lead author highlighted that drug discovery could become much easier with the help of the AI model.

The AlphaFold 3 is trained on a vast amount of research material and datasets about protein structures and their interaction with other molecules. By understanding the context and logic of protein structures, the LLM can predict how certain target zones will react when they come in contact with certain molecules.